Laser Cooling: Fundamental Properties and Applications

Laser Cooling
Pan Stanford Series on Renewable Energy — Volume 2
Laser Cooling
Fundamental Properties and Applications
edited by
editors
Preben Maegaard Galina Nemova
Anna Krenz
Wolfgang Palz
The Rise of Modern Wind Energy
Wind Power
for the World
Published by
Pan Stanford Publishing Pte. Ltd.
Penthouse Level, Suntec Tower 3
8 Temasek Boulevard
Singapore 038988
Email: editorial@panstanford.com
Web: www.panstanford.com
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
Laser Cooling: Fundamental Properties and Applications

Copyright © 2017 by Pan Stanford Publishing Pte. Ltd.
All rights reserved. This book, or parts thereof, may not be reproduced in any form
or by any means, electronic or mechanical, including photocopying, recording
or any information storage and retrieval system now known or to be invented,
without written permission from the publisher.
For photocopying of material in this volume, please pay a copying fee through
the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923,
USA. In this case permission to photocopy is not required from the publisher.
ISBN 978-981-4745-04-8 (Hardcover)
ISBN 978-981-4745-05-5 (eBook)
Printed in the USA

Contents
Preface xiii
1. Laser Cooling of Dense Gases by Collisional

Redistribution of Radiation 1
Anne Saβ, Stavros Christopoulos, and Martin Weitz
1.1 Introduction 1
1.2 Redistribution of Radiation 3
1.2.1 Basic Principle 3
1.2.2 Cooling Principle 4
1.3 Experimental Setup and Methods 7
1.3.1 High-Pressure Cells 8
1.3.2 Spectroscopic Setup 9
1.3.3 Thermal Deflection Spectroscopy 10
1.4 Laser Cooling Experiments on Dense Mixtures 14
1.4.1 Fluorescence of a Dense
Rubidium–Argon Mixture 15
1.4.2 Cooling of Dense Rubidium–Argon
Mixtures 17
1.4.3 Experiments on Potassium–Argon
Mixtures 21
1.5 Kennard–Stepanov Experiments 23
1.5.1 The Kennard–Stepanov Relation 23
1.5.2 Kennard–Stepanov Spectroscopy 25
1.5.3 Prospects of the Kennard–Stepanov
Technique 29
1.6 Conclusions 30
2. Laser Cooling in Rare Earth–Doped Glasses and Crystals 37

Galina Nemova
2.1 Introduction 37
2.2 Thermodynamics of Laser Cooling of Solids 40
2.2.1 Entropy for Near-Monochromatic
Radiation Flow 42
vi Contents
2.2.2 Conversion Efficiency 43

2.3 Fundamentals of Laser Cooling in Rare
Earth–Doped Solids 43
2.3.1 The Two-Level Model of Laser
Cooling in Rare Earth–Doped Hosts 43
2.3.2 Reabsorption or Fluorescence
Trapping and Total Internal Reflection 50
2.3.3 Cooperative Emission 53
2.3.4 Achievable Temperature 53
2.4 Optical Cavities 54
2.4.1 Intracavity Cooling 54
2.4.2 Nonresonant Cavity 56
2.4.3 Resonant Cavity 57
2.5 Laser Cooling in Rare Earth–Doped
Glasses and Crystals 57
2.5.1 Laser Cooling in Rare Earth–Doped
Glasses 60
2.5.1.1 Rare earth–doped fluoride
glasses 60
2.5.1.2 Laser cooling in rare
earth–doped fluorochloride
glasses: Yb3+:CNBZn and Er3+:
CNBZn bulk samples 63
Crystals 64
2.5.2.1 Yb3+-doped oxide crystal
(Yb3+:YAG) 64
2.5.2.2 Er3+-doped chloride
crystal (Er3+:KPC) 65
2.5.2.3 Rare earth–doped fluoride
crystals 66
Oxyfluoride Glass Ceramic 69
2.6 Temperature Measurements 69
2.6.1 Contact Technique 70
2.6.1.1 Thermocouples 70
2.6.1.2 Fiber Bragg grating 70
2.6.2 Noncontact Technique 70
Contents vii
2.6.2.1 Photothermal deflection 70

2.6.2.2 Thermal cameras 71
2.6.2.2 Differential luminescence
thermometry 71
2.6.2.3 Two-band differential
spectral metrology 72
2.6.2.4 Differential infrared
thermography 72
2.6.2.5 Polarization-resolved
thermometry 72
2.6.2.6 Mach–Zehnder
interferometry 73
2.7 Conclusions 73
3. Progress toward Laser Cooling of Thulium-Doped

Fibers 83
Nai-Hang Kwong, Rolf Binder, Dan Nguyen,
and Arturo Chavez-Pirson
3.1 Introduction 83
3.2 Theoretical Developments 85
3.2.1 Modeling the Cooling of Thulium
Atoms 85
3.2.2 Cooling Power of a Pumped, Doped
Fiber in Free Space 89
3.2.3 Heat Transport in a Structure
Cooled by a Laser-Cooled Fiber 93
3.2.3.1 Model of the cooling device 94
3.2.3.2 Numerical simulation of
heat flow in the cooling
device model 95
3.3 Experimental Developments 98
3.3.1 Fabrication of Highly Purified
Tm-Doped Tellurite Glasses 98
3.3.2 Design and Fabrication of Cooling
Fibers 100
3.3.3 Experimental Demonstration of
Optical Cooling in Fibers 102
3.3.3.1 Thermistor measurements 102
viii Contents
3.3.3.2 Noncontact measurements 105
4. Laser Cooling of Solids around 2.07 Microns:

A Theoretical Investigation 111
Guang-Zong Dong and Xin-Lu Zhang
4.1 Introduction 111
4.2 Conventional Laser Cooling of
Holmium-Doped Fluoride Crystals 113
4.3 Efficient and Enhanced Holmium Optical
Refrigeration via Co-Pumping 123
4.4 Energy Transfer–Enhanced Laser
Cooling in Rare Earth–Co-Doped Solids 133
4.5 Conclusions 141
5. Optically Cooled Lasers 147

Steven R. Bowman
5.2 Laser Thermal Management 148
5.3 Theory of Low–Quantum Defect Laser
Materials 151
5.3.1 Lossless Laser Materials 152
5.3.2 Optimization of Laser Material
Performance 157
5.3.3 Nonideal Laser Materials 163
5.4 Low–Quantum Defect Laser Experiments 168
5.4.1 Introduction to Experiment Review 168
5.4.2 Ytterbium Laser Experiments 169
5.4.3 Alternative Low–Quantum Defect
Laser Materials 177
5.5 Summary 178
5.6 Appendix 180
6. Methods for Laser Cooling of Solids 189

Stephen C. Rand
6.2 Cooling with Anti-Stokes Fluorescence 191
6.3 Brillouin Cooling 194
6.4 Raman Cooling 199
Contents ix
7. Deep Laser Cooling of Rare Earth–Doped Crystals

by Stimulated Raman Adiabatic Passage 211
Andrei Ivanov, Yuriy Rozhdestvensky,
and Evgeniy Perlin
7.2 Vibronic Model of Laser Cooling of Rare
Earth Ions 214
7.2.1 Vibronic Interaction 216
7.2.2 Model Hamiltonian for the Rare
Earth Ion System 216
7.2.3 Time Evolution of the Five-Level
System 218
7.3 Laser Cooling of the Yb3+:CaF2 System 222
7.3.1 Number of Electrons Involved
in the Cooling Cycle 224
7.3.2 Estimation of Cooling Performance 228
7.4 Conclusions 235
8. Bulk Cooling Efficiency Measurements of Yb-Doped

Fluoride Single Crystals and Energy Transfer–Assisted
Anti-Stokes Cooling in Co-Doped Fluorides 239
Azzurra Volpi, Alberto Di Lieto, and Mauro Tonelli
8.2 Optical Cooling Model for Rare Earth–
Doped Materials 241
8.3 Experimental Setup 247
8.3.1 Spectroscopic Setup 247
8.3.2 Cooling Setup 249
8.4 Experimental Results 251
8.4.1 Investigation of Anti-Stokes
Cooling in YLF Single Crystals
with Varying Yb Doping Levels 252
8.4.2 Yb–Tm Energy Transfer–Enhanced
Anti-Stokes Efficiency in Yb-Doped
YLF Single Crystals 260
8.5 Concluding Remarks 269
x Contents
9. Interferometric Measurement of Laser-Induced

Temperature Changes 273
B. Rami Reddy
9.1 Historical Development of Temperature
Sensors 273
9.1.1 Optical Sensor Technologies 274
9.1.2 Transmission Spectral
Measurements for
High-Temperature Measurement 275
9.1.3 Interferometric Sensors 275
9.2 Rare Earth Luminescence Spectroscopy
and Temperature Sensors 277
9.2.1 Differential Luminescence
Measurement 280
9.2.2 Yb3+-Doped YAG Spectra 280
9.3 Mach–Zehnder Interferometer 282
9.3.1 Theory of Mach–Zehnder
Interferometer 284
9.4 Optical Heterodyne Technique 289
9.4.1 Principle of the Optical Heterodyne
Detection Technique 289
9.4.2 Application of Optical Heterodyne
Detection Technique to Laser
Cooling/Heating of Solids 290
9.5 Michelson Interferometer 294
9.5.1 Theory of Michelson Interferometer 296
10. Fluoride Glasses and Fibers 305

Mohammed Saad
10.2 Bulk Glass 307
10.2.1 Glass Preparation 310
10.3 Glass Characterization 313
10.3.1 Differential Scanning Calorimetry 313
10.3.2 Glass Transmission 316
10.3.3 Refractive Index and Dispersion 317
10.4 Optical Fiber 321
10.4.1 Fiber Drawing 322
10.4.2 Fiber Attenuation 325
Contents xi
10.4.3 Mechanical Properties 328

10.4.4 Rare Earth–Doped Fibers 331
11. Crystal Growth of Fluoride Single Crystals for

Optical Refrigeration 337
11.2 The Czochralski Growth Method 343
11.3 Crystal Growth of Fluorides 347
11.4 The Growth Facility at the NMLA
Laboratory of Pisa University 350
11.5 Crystal Growth of Cooling Materials
at the NMLA Laboratory: Recent Results 355
11.5.1 The YLF Crystal 356
11.5.2 The Growth of Yb-Doped YLF
Single Crystals 358
11.5.2.1 The phase diagram of
the LiF–YF3 system 358
11.5.2.2 The growth of YLF single
crystals with varying Yb
doping levels 359
11.5.2.3 Preparation of crystal
seeds for the growth of
Yb-doped YLF single crystals 364
11.5.2.4 The growth of Yb-doped
YLF single crystals with
controlled Tm doping 365
11.5.3 Investigation of Sample Quality:
Structural Analysis, Testing of
Internal Defects, and Major
Contaminants 366
11.5.3.1 Structural analysis: X-ray
backscattering diffraction 366
11.5.3.2 Scattering analysis for
crystal defect detection 369
11.5.3.3 UV-NIR absorption
measurement for
detection of major
contaminants 370
xii Contents
11.5.4 Sample Fabrication 373

12. Microscopic Theory of Optical Refrigeration of

Semiconductors 377
Rolf Binder and Nai-Hang Kwong
12.2 Theoretical Foundation and the
Importance of Excitonic Effects 379
12.3 Effect of Luminescence Propagation
and Re-Absorption 386
12.4 Finite Spatial Beam Profiles 392
12.5 Theory of Passivation Layers 397
13. Coulomb-Assisted Laser Cooling of Piezoelectric

Semiconductors 409
Iman Hassani Nia and Hooman Mohseni
13.2 Piezoelectricity in Semiconductors 413
13.3 Basics of Optomechanical Cooling and
Amplification 416
13.4 Coulomb Interaction in Semiconductors
and Its Application for Laser Cooling 421
13.5 Effect of the Separated Electron and Hole
Wave Functions on the Recombination Rates 424
13.6 Formalism of Coulomb-Assisted Laser
Cooling of Semiconductors 430
13.7 Comparison of Coulomb-Assisted
Cooling in Piezoelectric Materials with
Collisionally Aided Laser Cooling 441
Index 453
Contents xiii
Preface
Refrigeration or cooling is a process in which a substance is

maintained at a temperature below that of its surroundings. It is
based on controlled heat removal from one location to another and
dates back thousands of years when people tried to find different
ways of preserving their food. At that time ice or snow, which
provided refrigeration through structures as simple as holes in the
ground or cold cellars, was the only available cooling technique.
In more recent history the efforts of scientists of the nineteenth
century to liquefy the permanent gases brought cryogenics into
existence. Cryogenics is the science of low temperatures, which
are below the point at which permanent gases begin to liquefy.
The invention of a laser was one of the groundbreaking scientific
achievements of the twentieth century, which further revolutionized
cryogenics. The rapid development of cryogenic techniques led to
a number of fundamental discoveries such as superconductivity
(Heike Kamerlingh Onnes, 1908), superfluidity (Pyotr Leonidovich
Kapitsa, John Frank Allen, and Don Misener, 1937), and Bose–
Einstein condensation (Eric Allin Cornell, Carl Edwin Wieman, and
Wolfgang Ketterle, 1995).
Today cryogenics continues to be a burning problem from
fundamental and applied points of view. Cooling with light, also
known as laser cooling or optical refrigeration, is one of promising
cryogenics techniques. Laser cooling is a physical process by which
a system loses thermal energy as a result of its interaction with laser
light. This process can be based on the Doppler effect, which can be
used to trap atoms and slow their motion to cool a substance. This
method along with further evaporative cooling has permitted to
realize Bose–Einstein condensation of atomic and molecular species.
Optical refrigeration can also be based on redistribution of radiation
in dense gas mixtures. Both these techniques permit to remove the
thermal energy contained in translational degrees of freedom. They
are suitable for gases because at low temperatures the thermal
energy in gases is primarily contained in translational degrees of
freedom. In some systems, for example, in solids, the thermal energy
is primarily contained in vibrations (phonons). Such systems can
xiv Preface
be laser cooled with anti-Stokes fluorescence. In that case a system

illuminated with laser light at long wavelength tail of its absorption
spectrum emits photons of higher energy than the absorbed ones as
a result of phonon absorption in the system. Laser cooling of solids
was first experimentally demonstrated in 1995 with ytterbium-
doped fluoride glass. Since the first experimental observation laser
cooling of solids has been realized with Yb3+, Er3+, and Tm3+ ions
doped in a wide variety of low phonon glasses and crystals. Today’s
record temperature achieved in a Yb3+:YLF sample is 91 K, which is
far below the temperature that can be achieved with any other solid
state cooling techniques. New materials as well as old ones with
extremely low levels of impurities can permit to reach even lower
temperatures in the nearest future. Direct band-gap semiconductors
are very promising candidates for laser cooling. Laser cooling is
progressing rapidly. The latest achievements have motivated me to
pull together a team of researchers who are doing frontier research
in the field of laser cooling to discuss the fundamental principles and
applications of this promising field of physics. Their research and
experience have been organized into the 13 chapters of the book.
In Chapter 1 Anne Saß, Stavros Christopoulos, and Martin Weitz
review their recent results on laser cooling by redistribution of
radiation in dense gas mixtures. The achieved cooling powers in
these experiments are several orders of magnitude larger than in
the Doppler cooling experiments for dilute atomic vapors, and the
typically achieved cooling efficiencies are in the order of several
per cent. The cooling results are presented in the form of thermal
deflection spectroscopy measurements on both rubidium– and
potassium–argon mixtures. They also present the Kennard–Stepanov
relation, connecting absorption and emission profiles, and apply it
to the dense gaseous ensembles under investigation. The authors of
the chapter discuss their future experimental aims that include the
study of homogeneous nucleation in saturated vapor as well as the
application of this very technique to molecular–rare gas ensembles.
In Chapter 2 Galina Nemova presents a detailed analysis of laser
cooling with anti-Stokes fluorescence in a two-level rare earth–
doped solid state system. The influence of all cooperative effects in
rare earth–doped samples on the laser cooling process is considered.
Thermodynamics of laser cooling is discussed in the chapter. The
Preface xv
advantages of different host materials with low phonon energy

for laser cooling with anti-Stokes fluorescence are considered.
In this chapter one can see how the efficiency of the laser cooling
process can be enhanced with optical cavities. The author presents
an overview of techniques, which can provide a local temperature
measurement as well as the bulk temperature of a sample. All these
techniques have some advantages and disadvantages, which are
discussed in the chapter.
Chapter 3 is intended to give an overview of some theoretical
and experimental efforts aimed at developing an optical cooling
system based solely on optical fibers. In this chapter Nai-Hang
Kwong, Rolf Binder, Dan Nguyen, and Arturo Chavez-Pirson show
that such a cooling system has unique advantages, including the
possibility for dumping heat waste at arbitrary remote locations
and the light-weight, compact, and robust nature of the system.
The theoretical basis includes a rate equation–based cooling model
for the active ions in the doped fiber as well as heat transport
simulations. Experimental results for luminescence spectra for
various pump powers show promising trends when compared with
spectra as a function of temperature, indicating the possibility of net
fiber cooling.
In Chapter 4 Guang-Zong Dong and Xin-Lu Zhang propose a mod-
ified excitation cooling scheme based on ground-state absorption of
5I Æ 5I at around 2.07 μm and excited-state absorption of 5I Æ 5F
8 7 5 5
at around 2.28 μm in Ho3+ ions. In this scheme large amount of heat
produced by nonradiative 5I5 Æ 5I6 and 5I6 Æ 5I7 transitions in the
conventional anti-Stokes cooling case can be avoided, and both the
cooling efficiency and the cooling power can be prompted. They also
propose to improve the cooling performance of Ho3+ optical refrig-
eration by co-doping Tm3+ and Ho3+ ion in the host.
Since the invention of the laser, engineers have searched for new
ways to increase the laser’s power and brightness. These efforts have
largely been focused on improved ways to remove the waste heat
generated in the lasing medium. In Chapter 5 Steven R. Bowman
reviews a novel approach to the problem of laser power scaling:
minimizing the waste heat through the incorporation of anti-Stokes
fluorescence cooling. This approach has been shown to significantly
reduce heat generation in carefully controlled laser systems. This
xvi Preface
chapter reviews the theory and practice of this novel mode of laser
operation.
In Chapter 6 Stephen C. Rand compares three conceptually
different approaches to laser cooling of solids. He offers the unique
perspective that in solids where free translational motion is absent
there are nevertheless interactions analogous to the Doppler effect
that may be used to address the energy and momentum of collective
excitations of the medium. Moreover there are reasons to explore
them systematically. Two methods of cooling solids in particular,
which are reminiscent of interactions with free atoms, are analyzed
from this perspective to examine their detailed advantages and
disadvantages. While attempting to profit from the earlier history of
laser cooling in gases and to provide points of comparison between
different techniques, this chapter identifies an approach to cooling
that results in a temperature-independent cooling rate.
In the “traditional” model, which describes optical refrigeration
of rare earth–doped solids based on anti-Stokes fluorescence, a
cooling cycle is provided by resonance transitions between the upper
ground and lower excited levels of ions, phonon thermalization,
and subsequent radiative relaxation to the ground states. In such a
model, the rate of the cooling cycle is low because of the small cross
section of forbidden transitions between the levels of the same ion
configuration and the low population of the upper ground level. In
Chapter 7 Andrei Ivanov, Yuriy Rozhdestvensky, and Evgeniy Perlin
consider a cooling process with stimulated Raman adiabatic passage
for optical pumping of a rare earth–doped crystal taking into account
vibronic interaction at the Stark split sublevels of ion manifolds.
They show that coherent pumping by stimulated Raman adiabatic
passage leads to the deep laser cooling of rare earth–doped crystals
that have technologically attainable purity to date.
In Chapter 8 Azzurra Volpi, Alberto Di Lieto, and Mauro Tonelli
report on cooling efficiency measurements performed on various
Yb3+-doped fluoride single crystals grown in their laboratory. They
show that the investigation of the cooling process for increasing Yb3+
concentrations in LiYF4 (YLF) single crystals results in improved
cooling efficiency via significant decrease of detrimental parasitic
absorption. Energy transfer processes between Yb3+ ions and low
concentrations of rare earth ions impurities are investigated in this
chapter to analyze the effect of detrimental impurities, which partic-
Preface xvii
ipate in multi-phonon relaxation of excited ions. It is demonstrated

that Yb3+–Tm3+ controlled co-doping led to observation of energy
transfer–enhanced Yb3+ anti-Stokes efficiency. The results achieved
with Yb3+–Tm3+ co-doped samples are presented in this chapter.
In Chapter 9 B. Rami Reddy demonstrates the use of Michelson
and Mach–Zehnder interferometers to measure the sample temper-
ature changes due to laser beam heating. He has developed an ana-
lytical model to estimate the temperature change that depends on
fringe (or wave) counting. The passage of a 300 mW of 915 nm laser
beam through 1.54 cm long Yb3+:YAG sample raised its temperature
by 7.1 ± 0.8 K. The passage of a 115 mW focused argon-ion beam
through a 1 cm long Pr3+:YAG sample raised its temperature by 11.7
± 1.0 K. Optical interferometric measurement of temperature chang-
es agreed with that measured by a thermocouple. These methods
apply equally well to estimate the temperature changes in laser cool-
ing of solids also. He has also developed theoretical formulations for
the use of optical heterodyne technique in laser cooling studies. His
theory reveals that sample temperature changes as small as 1.4 mK
can be measured with the heterodyne technique.
Optical fibers are now critical parts in many high-tech devices.
Of course fiber parameters have to be tailored to each application
needs. Unfortunately, there is no single fiber material that can
fulfill all types of application needs, but there are some important
criteria that one has to consider when choosing an optical fiber to
be used in a specific application. The first criterion is technical. The
second criterion is economical. The technology should be mature
enough to provide high-quality optical fibers at effective cost with
a reasonable yield. Fluoride glasses are very promising hosts for
laser cooling. In Chapter 10 Mohammed Saad presents the latest
technological achievements concerning three main glass families
as representatives of fluoride glasses in general: fluorozirconate,
fluoroaluminate, and fluoroindate.
In the last years, the advances enabled by Yb3+-doped fluorides
have tremendously increased the interest in this technology.
Cryogenic operations as well as competitive steady-state cooling
powers have been achieved in the temperature range below 250 K.
Minimum achievable temperatures are approaching 80 K. Efficient
cooling performances, however, are firstly related to the optical
quality of active materials. Stringent requirements on the purity and
xviii Preface
structural quality need to be satisfied in order to achieve net bulk

cooling. In Chapter 11 Azzurra Volpi and Mauro Tonelli consider
growth techniques and methods for optical and structural quality
assessments of fluoride single crystals for optical refrigeration.
In Chapter 12 Rolf Binder and Nai-Hang Kwong give an overview
of some theoretical efforts to analyze optical refrigeration of
inorganic semiconductor crystals. A fermionic theory, describing the
conduction and valence bands, combined with Coulomb interaction
between the excited charge carriers yields the absorption and
luminescence spectra that include excitonic effects valid for
arbitrary excitonic ionization ratios. The microscopic theory used
as an input to Sheik-Bahae and Epstein’s semiconductor cooling
model yields a cooling analysis that has been evaluated for GaAs.
At low temperatures excitonic effects are found to be crucial for
cooling to be possible. Extensions of the basic cooling model for bulk
crystals include light propagation and luminescence re-absorption
effects and the effects of passivation layers in doped, spatially
inhomogeneous structures. While the numerical results presented
here are restricted to GaAs, the general theoretical formulation is
valid for any direct-gap crystalline semiconductor.
In Chapter 13 Iman Hassani Nia and Hooman Mohseni discuss the
mechanisms that can potentially increase the laser cooling efficiency
to make it practical for bulk semiconductors. They consider Coulomb
interaction in piezoelectric semiconductors. The authors describe
both Coulomb interaction and piezoelectricity separately to establish
the theoretical bases for understanding Coulomb-assisted laser
cooling. Their approach is inspired by the optomechanical cooling of
micro- and nano-mechanical systems since each phonon mode acts
like a mechanical mode of vibration with a low quality factor.
I would like to sincerely express my gratitude to all the
contributing authors for their excellent work and patience during
the preparation and editing of this book in the International Year of
Light 2015.
Galina Nemova
Montreal, Canada
Summer 2016
Chapter 1
Laser Cooling of Dense Gases by

Collisional Redistribution of Radiation
Anne Saβ, Stavros Christopoulos, and Martin Weitz

Institut für Angewandte Physik der Universität Bonn, Wegelerstr. 8,
53115 Bonn, Germany
sass@iap.uni-bonn.de, christopoulos@iap.uni-bonn.de, weitz@iap.uni-bonn.de
1.1 Introduction
Light as a tool for cooling matter was considered already in 1929
by P. Pringsheim [1]. The experimental realization of the laser [2]
paved the way for experiments on different laser cooling techniques.
Among these, Doppler cooling of dilute atomic gases is probably the
most common [3–5], first successfully investigated in the 1980s [6].
As an important step, this technique along with further evaporative
cooling allowed for the realization of Bose–Einstein condensation of
different atomic [7, 8] and molecular species [9] and has today grown
into a robust research area. An alternative approach for cooling is
anti-Stokes cooling in multilevel systems, culminating, for example,
in the cooling of solids. For the first time suggested in 1950 [10] and

Edited by Galina Nemova
Copyright © 2017 Pan Stanford Publishing Pte. Ltd.
ISBN 978-981-4745-04-8 (Hardcover), 978-981-4745-05-5 (eBook)
www.panstanford.com
2 Laser Cooling of Dense Gases by Collisional Redistribution of Radiation
experimentally realized in 1981 [11], this technique leads nowadays

to a significant cooling of heavy metal fluoride glass doped with, for
example, trivalent ytterbium ions [12, 13].
In this chapter, we review results on laser cooling by redistribu-
tion of radiation in dense gas mixtures. Redistribution of fluores-
cence is a well-known effect observed in experiments of magneto-
optically trapped atoms, where it acts as the main loss mechanism
of the trap [5]. Considering atomic collisions at room tempera-
ture, redistribution of fluorescence is a consequence of collisional
aided excitation [14]. In a theoretical work in 1978, P. Berman and
S. Stenholm proposed a cooling mechanism based on collisionally
aided fluorescence and the related energy loss in a two-level system
[15]. Corresponding experiments with gases at moderate densities
never reached the cooling regime [16]; only heating for blue-de-
tuned excitation was observed for these conditions.
Using a high-pressure environment with a system of
rubidium atoms subject to 230 bar of argon buffer gas, our group
experimentally demonstrated laser cooling by redistribution of
fluorescence in 2009 [17]. At the used buffer gas pressures of a few
hundred bars, the optical transitions are broadened to linewidths
that are in the same order of magnitude as the thermal energy, kBT,
in frequency units, where kB is the Boltzmann constant and T the
temperature of the gas mixture. To obtain an idea of the cooling
principle in this regime, assume the formation of transient, alkali–
noble gas quasi-molecules during each binary collision of the two
species. The atomic resonances are thus perturbed by means of the
rising intermolecular potential, allowing for absorption of far red-
detuned incident radiation. For typical parameters, the radiative
lifetime of the excited state exceeds the time of such a collision by 3
to 4 orders of magnitude, which is a few nanoseconds compared to
picoseconds.
Subsequent decay of the excited state occurs mostly at larger
interatomic distances where the alkali resonance frequency is
close to its unperturbed value. In this manner, the mean emitted
fluorescence has a smaller wavelength than the absorbed photon;
thus energy in the order of the thermal energy is extracted from the
sample, cooling the dense mixture.
In the following sections, we initially discuss the mechanism of
redistribution of radiation in a more detailed way (Section 1.2) and
subsequently present the experimental setup and the high-pressure

chambers in Section 1.3. Experimental results will be reviewed
throughout Sections 1.4 and 1.5, before concluding and giving an
outlook in Section 1.6.
1.2 Redistribution of Radiation

Before discussing the cooling principle for the case of alkali–noble
gas mixtures in more detail, we give a brief overview of the basic
principle of redistribution of radiation.
1.2.1 Basic Principle

Collisionally induced redistribution of radiation is based on the
interaction between an incident light field with an optically active
atom of species A surrounded by perturbing atoms of species B.
Binary collisions between atoms of the two species can lead to
broadening of atom’s A resonances [18, 19]. This enables absorption/
emission of radiation at a wavelength nonresonant to the free and
unperturbed atom. The effect has been observed for the first time in
strontium vapor at low buffer gas pressures [20], while a theoretical
prediction can be found in Refs. [21, 22].
Let us consider photons with energy ℏωL. We will further
consider that the difference between the transition energy ℏω0
and the photon energy is small compared to the thermal energy
kBT of the perturbers. In this case, the photons are either scattered
elastically in terms of Rayleigh scattering, or they are absorbed
and reemitted as fluorescence (frequency ωFl) if the energy gap is
overcome by energy transfer from the perturbing atoms. A schematic
of the described process is shown in Fig. 1.1. Note that in the case of
a multilevel atom A surrounded by perturbing atoms, it is possible
to achieve excitation into levels that would be too far off resonant in
the undisturbed case.
If we consider the collisional pair (A – B) to form a quasi-molecule
at small interatomic distances, we can intuitively understand the
redistribution process. When the collision takes place, the atom A
absorbs the incident photon. The quasi-molecule is thus excited
from the ground state into the excited state. Following the collision,
the distance between the collisional partners grows and after its 1/e
natural lifetime—which is of the same order as in the unperturbed
case [23]—the electronically excited state decays under the
emission of a fluorescence photon. Since the energy levels at larger
interatomic distances approach those of the unperturbed radiating
atom A, the frequency of the emitted photon ωFl will be close to the
resonance frequency of atom A.
Figure 1.1 Process of redistribution of radiation for a two-level atom A

surrounded by perturbing atoms of species B (left). Depending
on the degree of elasticity of the collision with the perturbing
atoms, resonant Rayleigh scattering (ω = ωL) or fluorescence at
a different frequency ωFl can be detected (right).
It can be shown experimentally that the intensity of the

redistributed fluorescence is proportional to the intensity of the
incident light field and to the number density of species B at a given
(far) detuning in case of low intensities and pressures. For larger
intensities and number densities, the emitted fluorescence will at
some point exhibit saturation [18].
1.2.2 Cooling Principle

We conduct experiments with alkali atoms subject to a high-
pressure noble gas. With regard to the nomenclature used above,
the alkali atoms are treated as species A, while the noble gas atoms
as the perturbing species B. In these experiments, considerable
broadening of the alkali resonances occurs due to frequent collisions
with the noble gas atoms, making the absorption of far-red-detuned
radiation feasible. As already noted above, the radiative lifetime
of the electronically excited quasi-molecular state is similar to the
lifetime of the excited alkali state. Our main experimental focus

lies on rubidium atoms subject to high-pressure argon gas. The
corresponding quasi-molecular potentials can be found in Ref. [24],
calculated out of [Rb+] and [e− − Ar] pseudo-potentials. The resulting
cooling principle is sketched in Fig. 1.2 for the case of rubidium
atoms subject to argon buffer gas.
Figure 1.2 Cooling principle for the case of rubidium atoms subject to
high-density argon buffer gas, based on the calculated pseudo-
potentials [24]. If a rubidium atom collides with an argon atom,
absorption of far red-detuned radiation becomes feasible.
The formed quasi-molecule is excited, and the excited state
subsequently decays under emission of a fluorescence photon
mostly at larger interatomic distances, where the energy
levels correspond to those of the unperturbed alkali atom.
The fluorescence frequency ωFl is larger than the frequency of
the absorbed photon: an amount of energy in the order of the
thermal energy, kBT, is extracted from the sample. A cooling of
the gas mixture occurs.
For small interatomic distances, a photon can be absorbed

also for red detunings of several nanometers. After its 1/e natural
lifetime, the upper electronic state decays emitting a fluorescence
photon. Since the mean frequency of the detected fluorescence is
larger than the frequency of the absorbed photon, a cooling of the
dense gas mixture is expected as energy is extracted.
Our choice of rubidium atoms as the optically active species

becomes now clear. The large elasticity of collisions between the
electronically excited states of the rubidium D-lines and noble gas
atoms has long been proven [25]. The quasi-molecular potential
has a depth of the order of kBT, allowing for a significant part of the
atoms’ kinetic energy to be extracted. Finally, the cooling transitions
are in the near-infrared spectral regime, making them conveniently
optically accessible to commercial laser systems.
The cooling power Pcool from these experiments can be
estimated in a similar way as for anti-Stokes cooling [26, 27] from
the mean fluorescence frequency νfl, the incident power Popt, and the
absorption probability of the gas mixture at incident frequency ν,
α(ν).
n fl - n (1.1)
Pcool = a (n )Popt
n
The typical energy that can be extracted from the sample per
cooling cycle is of order of the thermal energy kBT. In the experiments
performed so far, we observe cooling powers Pcool in the order of
up to 100 mW [17, 23, 28, 29]. Note that it is important to choose
a detuning with a high-enough absorption probability a(ν). We are
aware that the collisional broadening plays a crucial role here. A
measurement of typical fluorescence data recorded with the dense
gas mixture will be shown in Section 1.4.1.
Experimentally, it turns out to be crucial that the used buffer
gas is of high purity to suppress the influence of concurring
heating effects, of which quenching is the most known. In the case
of quenching, the excitation energy is transferred to an atom of a
third species polluting the dense gas mixture when colliding with
the quasi-molecule. Relaxation can then occur via nonradiative
decay, and so energy is kept in the system, heating the gas. To reduce
possible quenching, an effect known to occur, for example, from
an admixture with nitrogen, we use a buffer gas with high purity,
namely argon gas preferably with a purity of 99.9999 (argon 6.0).
Residual heating can also be caused by energy pooling. In this
scenario, during the collision of two excited alkali atoms one of them
gets further excited. Its relaxation to the ground state implicates
the emission of a fluorescence photon with higher energy. For the
case of rubidium atoms, the reaction equation of that process can be
written as [30]:
Experimental Setup and Methods 7
mv12 mv22 mv2 mv22

Rb* + Rb* + + Æ Rb + Rb** + 1 + . (1.2)
2 2 2 2
The sum of the energies of the single excited rubidium atoms
before the collision is higher than the energy of the twice excited
atom. The surplus of the energy is thus kept in the system as kinetic
energy of the collisional partners [31]. For rubidium atoms subject
to a high-pressure noble gas, fluorescence originating from the
6p Æ 5s transition then occurs, and can be detected as blue light at a
wavelength λ ≈ 420 nm in the experiments.
At room temperature, the alkali atoms are in the solid phase.
To produce a sufficient alkali vapor number density the system is
heated to temperatures of up to 680 K, while the buffer gas pressure
ranges between 50 bar and 230 bar. Under these conditions, we
calculate that the dense gas ensemble consists of approximately
1017 cm–3 alkali atoms and 1021 cm–3 noble gas atoms. Then, we can
estimate the mean free path length λfp for a gas of colliding atoms
(cross section σ) and number density n
1
lfp = (1.3)
ns
Let us consider the geometric cross section πd2, where d denotes
the diameter of a rare gas atom, which is twice the van der Waals
radius (d = 2rvdW = 3.76 × 10−10 m for argon [32]). Then, for argon
pressure of 200 bar, the mean free path length is in the order of only a
few nanometers, λfp ≈ 2.3 nm. Following the Boltzmann distribution,
still considering a temperature T = 680 K, one can also estimate the
collisional rate to 2.4 × 1011/s. Comparing this to the natural lifetime
of the first excited alkali atom level (27 ns), we calculate that 104
collisions occur within this period of time.
1.3 Experimental Setup and Methods

In the following section, we present our experimental setup in detail,
including the techniques used to measure atomic fluorescence and
temperature changes inside the gas mixture. We offer insights into
the high-pressure cells employed (Section 1.3.1), as well as into the
spectroscopic techniques (Section 1.3.2) and the thermal deflection
setup (Section 1.3.3).
1.3.1 High-Pressure Cells

For the creation of a high-pressure regime, two different designs of
optical cells are employed. Initial experiments were performed with
homebuilt cells with volumes of just a few cubic centimeters. The
optical access is provided by sapphire windows, which are sealed
with Inconel® c-rings into the steel body of the cell and tightly
flanged. This design allows for a quick change of the optical windows
and, related to this, a change of the inner length in the cell due to the
special design we use here. A sketch of the homebuilt high-pressure
cell can be seen in Fig. 1.3.
Figure 1.3 Schematic of a high-pressure cell (homemade design). The

ampoule containing the alkali metal can be put into the
reservoir. Optical access is provided by two optical sapphire
windows opposite to each other. A manometer and a pressure
valve are mounted on the gas inlet to monitor and adjust the
pressure in the cell. High temperatures can be achieved by
wrapping heating wires (not sketched for better visibility)
around the cell body, enabling the creation of sufficient alkali
vapor density.
The second design is commercial, manufactured by SITEC-Sieber

Engineering AG. In this case, the sapphire windows are mounted
following a Bridgman sealing technique [33]. The different sealing
mechanism offers a better long-term stability of several months.
Buffer gas pressures of several hundred bar and temperatures of
up to 720 K can be achieved using both cell designs. A manometer,

placed on the gas inlet of each cell, allows for constant monitoring of
the pressure, ensuring that no leaks appear.
A process including ultrasound cleansing and a bake-out period
is followed in order to purify the cells, before an ampoule containing
1 g of alkali metal is placed in the reservoir. Subsequently, argon gas
is filled in, and the cells are heated by wrapping electrical heating
belts around them. In this manner, the alkali atoms are vaporized in
a high–buffer gas environment acquiring sufficient optical densities
that allow us to begin our experimental efforts.
1.3.2 Spectroscopic Setup

For the excitation of the dense mixture, we employ emission of a
continuous-wave Ti:sapphire ring laser that is guided through the
cell volume. This type of laser system also allows for an easy change
of the incident wavelength, exploited in energy dependence studies
of the fluorescence.
Figure 1.4 Fluorescence detection setup. The incident Ti:sapphire laser

beam is spatially filtered and tightly focused into the high-
pressure cell. In the back direction, atomic fluorescence is
collected and filtered and afterward detected by an optical
spectrum analyzer.
For most experiments, both the incident laser beam and the
emitted fluorescence were spatially filtered with confocal pinholes.
The incoming laser beam was strongly focused to a beam waist of
approximately 10 μm. Fluorescence was then collected from that
area and analyzed by an optical spectrum analyzer. A sketch of that
setup is depicted in Fig. 1.4.
1.3.3 Thermal Deflection Spectroscopy

To determine the temperature change in the dense mixture induced
by the cooling laser beam in a quantitative way, thermal deflection
spectroscopy [34–36] is employed. The basic idea behind this
technique is that a temperature change in the gas produces a
corresponding change of density, which in turn induces a variation
of the refractive index. This variation is investigated by irradiating a
second, nonresonant laser beam collinear to the cooling laser beam
into the dense mixture. The cooled gas acts like a graded index lens,
causing the probe laser beam to be deflected and consequently to
change its propagation direction. By measuring the deflection angle
as a function of distance between the two laser beams, the deflection
profile can be mapped. In the following, we deduce the relation
between the deflection angle and the temperature change according
to the work of Ref. [35]. A basic sketch of the concept is shown in
Fig. 1.5. Considering a Gaussian laser beam propagating through
a spatially varying profile of the refractive index, the change of its
propagation direction can be written as follows:

d Ê dr ˆ 
n = — ^ n(r , t ) (1.4)
dz ÁË 0 dz ˜¯
where r is the transversal displacement of the probe beam to its
original propagation direction, n0 denotes the refractive index of the

gas mixture at normal conditions, and — ^ n(r , t ) is the gradient of
the refractive index orthogonal to the propagation direction [37].
Integration of Eq. 1.4 leads to a relation of the (measurable)
deflection angle φ and the spatial change of the refractive index

dr
j= (1.5)
dz
1 
=
n0 Ú
— ^ n(r , t )dz (1.6)
1 dn
=
n0 Ú dr dz (1.7)
At this point we consider the deflection angle φ in radial
direction r to be small (dr/dz = tan φ ≈ φ). We derive an expression
for the temperature change as a function of the deflection angle by
reshaping Eq. 1.7:
1 dn
j=
n0 Ú dr dz (1.8)
1 dn dT
j=
n0 Ú dT dr dz (1.9)
Figure 1.5 Thermal deflection technique. The cooling beam and the
so-called probe beam are sent collinearly into the dense gas
mixture. The cooling zone induces changes to the gas density
and consequently to the refractive index. As a result, the
cooling zone acts like a graded index lens for the nonresonant
probe beam. By scanning the radial distance between the two
beams, the angular deflection φ can be measured as a function
of r.
Regarding Eq. 1.9, we can further simplify dn/dT in order

to obtain an expression that is only a function of experimental
parameters. According to the work of Spear et al. [36], dn/dT is
in general determined by two different mechanisms, the thermal
expansion and a thermal shift of the absorption bands of the media.
The choice of the nonresonant probe laser beam makes it feasible to
suppress the influence of absorption bands. We can simplify Eq. 1.9
by considering the Lorentz–Lorenz relation [38]
1 Ê n2 - 1 ˆ
= = const. (1.10)
r ÁË n2 + 2 ˜¯
where ρ is the density of the medium. Assuming that the dense gas
mixture behaves like an ideal gas, its refractive index n is close to
unity. The temperature derivative of Eq. 1.10 can then lead to the
following expression for dn/dT:
dn (n - 1) (1.11)
ª-
dT T
Combining Eqs. 1.9 and 1.11, we obtain
(n - 1) dT
j=- Ú T dr
dz (1.12)
To get a rough estimation for the analysis of the measured data,

we now consider the fraction (n − 1)/T in Eq. 1.12 to be constant
along the propagation direction z
(n - 1) dT (1.13)
j=-
T Ú
dr
dz
This assumption holds for small temperature changes compared

to the initial temperature of the dense gas mixture. If we further
assume that the cooling laser beam has a Gaussian intensity profile
and that the thermal energy is removed with a rate comparable to its
intensity distribution, the relative temperature change ΔT(r, z) can
be expressed as follows:
ΔT(r, z) = ΔT(r, 0)◊ e–αz (1.14)
Hence, as a function of experimental parameters, one derives for
ΔTexp:
•
T a e -a z
DTexp (r , z ) = ◊
(n - 1) (1 - e -a L ) Ú
j(r ¢ )dr ¢ (1.15)
r
where r denotes the transverse displacement between cooling and

probe laser beam, L is the inner length within the high-pressure cell,
T the initial temperature of the dense gas mixture, n is the refractive
index of the buffer gas under the particular experimental conditions
(p, T) for the wavelength of the probe laser beam, and α denotes the
absorption coefficient of the gas and is considered constant for the
analysis of the measured deflection data. In our present model, heat

transport is considered to occur only in the radial direction because
the absorption length, 1/α, is much longer than the laser focal radius
w0. The heat transfer is described by
— (κ—T) = –Q(r, z) (1.16)
where κ is the thermal conductivity of the buffer gas. The cooling
source Q(r, z) of a Gaussian laser beam can be expressed as a function
of the cooling power Pcool, the cooling beam focal radius w0, and the
absorption coefficient α [39, 40]:
Q(r , z ) µ a ◊ Pcool ◊ e -2r /w02

◊ e -a z (1.17)
2

Inspection of Eq. 1.16 indicates that the temperature change

inside the gas is a function of the thermal conductivity of the buffer
gas. Experiments with heavier rare gases such as krypton and
xenon are thus supposed to allow for larger temperature changes
compared to the so far mainly used argon. In the past, this has often
proved problematic as both krypton and xenon are hardly available
in purities of 6.0. We have nevertheless achieved first encouraging
results with both species as buffer gas [41]. For our experiments,
we use a He–Ne laser as a probe beam (wavelength λ ≈ 633 nm).
We are positive that this wavelength is far enough nonresonant with
the dense gas mixtures of our experiments (rubidium and argon/
helium/krypton/xenon as well as potassium–argon mixtures).
Furthermore, the choice of a probe laser wavelength in the visible
range strongly facilitates the alignment of the setup that can be seen
in Fig. 1.6. The Ti:sapphire laser beam is irradiated into the dense
gas mixture, while the collinear He–Ne laser beam can be laterally
moved by a beam splitter, allowing for a controllable variation of the
distance between the two beams. The angular deflection is measured
by a position sensitive photodiode behind the cell. To make sure that
only the probe beam is detected, an infrared filter is placed directly
in front of the photodiode. Finally, a mechanical chopper on the
cooling beam path ensures that angular deflection measurements do
not entail offsets, while simultaneously allowing for the observation
of deflection statistics at each lateral offset position.
In most of the experiments, the width of the refractive index
profile is scanned over a few millimeters in 50–100 μm steps.
Experiments are conducted with both collimated and focused
beams. In Fig. 1.6 the latter case is presented. The beam waist w0
of the probe beam is set to be smaller than that of the cooling laser
beam so that the entire cooling area can be resolved.
Figure 1.6 Experimental cooling setup incorporated with thermal deflec-

tion technique. A movable beam splitter allows for scanning
the lateral offset between the cooling beam (Ti:sapphire laser)
and the nonresonant probe beam (He-Ne laser). The angular
deflection is measured as a function of position by a position
sensitive diode. An infrared filter blocks the residual cooling
beam behind the cell, while a mechanical chopper allows for
measuring offset-free deflection statistics.
We discuss results on thermal deflection spectroscopy

measurements in Sections 1.4.2 and 1.4.3 for the case of rubidium
and potassium atoms both in dense argon buffer gas.
1.4 Laser Cooling Experiments on Dense

Mixtures
In this section, we lay focus on laser cooling experiments of dense
atomic mixtures. We begin by presenting a fluorescence spectrum
which exhibits a clear redistribution of radiation (Section 1.4.1),
before discussing thermal deflection spectroscopy measurements
on dense rubidium–argon ensembles (Section 1.4.2). We conclude
with the cooling results of dense potassium–argon mixtures (Section
1.4.3).
Laser Cooling Experiments on Dense Mixtures 15
1.4.1 Fluorescence of a Dense Rubidium–Argon Mixture

For the case of rubidium atoms subject to 160 bar argon atoms at
a temperature T = 580 K, fluorescence spectra are recorded for an
incident laser power of 100 mW at a wavelength λ ≈ 810 nm using
the setup described in Section 1.3.2. The recorded fluorescence as a
function of wavelength is presented in Fig. 1.7. A clear redistribution
of fluorescence toward the rubidium D resonances can be observed.
Residual scattered light at the incident wavelength is also detected
by the spectrometer.
Investigation of the atomic resonances leads to the conclusion
that they are both pressure and saturation broadened. The
linewidths of the two sharper resonances visible on top the broad
envelope are Δν¢(D1) = 3.4 THz, Δν¢(D2) = 5.9 THz, respectively, as
derived from Lorentzian fits. Early work for pressure broadening of
alkali atoms by noble gases at roughly similar noble gas pressures
has been performed by Shang-Yi [42].
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Figure 1.7 Recorded fluorescence of rubidium atoms subject to 160 bar

argon buffer gas. Both rubidium D transitions are spectrally
resolved. Compared to the unperturbed resonances D1, D2,
a strong shift is observed (center wavelengths labelled with
D2¢, D1¢, and dashed lines). Due to imperfect filtering, residual
scattered light of the incident laser beam at λ ≈ 810 nm is also
visible. A clear redistribution toward the D-line resonances is
observed.
The resonances are also shifted to higher wavelengths, λ(D1¢)

≈ 800 nm, λ(D2¢) ≈ 785 nm, when compared to the undisturbed
resonances at λ(D1) = 795 nm, λ(D2) = 780 nm [43]. The recorded
spectrum is asymmetric toward the center wavelength of the
fluorescence. The red wing of the broad envelope, in the region of
higher wavelengths, stretches over a significantly higher range than
the blue wing. This is in good agreement with a quasi-static approach
of the pressure-broadened rubidium argon quasi-molecules, where
the red wing is attributed to the A2P3/2 Æ X2S1/2 transition [44].
The spectrum presented clearly shows redistribution of radiation,
which is a prerequisite for the observation of cooling in the dense
gas mixture.
Figure 1.8 Calculated cooling power Pcool (represented by squares) as a

function of excitation energy, using the differential energy, (vfl
– v)/v, extracted directly from fluorescence spectra. The solid
line is used as a guide to the eye.
Further investigation involves recording a number of these

spectra using different excitation energies. In this manner, the
corresponding mean fluorescence νfl can be determined directly
from the spectra, and thus, an estimation of the cooling power
Pcool becomes feasible through Eq. 1.1. Corresponding results are
presented in Fig. 1.8, where the calculated cooling power (squares)
is presented as a function of the excitation energy [17]. We observe
that the value of zero detuning, corresponding to zero cooling power,

is located at 375 THz. Hence, we expect that for lower excitation
frequencies (red detunings), the dense gas mixture will be cooled,
while for higher (blue detunings) it will be heated. In the subsequent
sections we review results on the redistribution laser cooling, which
verify this prediction with small deviations attributed to additional
heating effects due to quenching.
1.4.2 Cooling of Dense Rubidium–Argon Mixtures

The addressed cooling transitions of rubidium and potassium are
spectrally close. As rubidium allows for a higher number vapor
density at a certain temperature compared to potassium [45, 46], we
initially focus on experiments with the heavier alkali species solved
in rare gases.
In the proof of principle experiment [17], a 2 W cooling beam
(diameter of the collimated beam was 2w0 ≈ 1 mm) at a wavelength
of λ ≈ 820 nm is irradiated into the high-pressure cell containing
rubidium at an initial temperature of 620 K under 230 bar of argon.
Initial evidence of cooling can be obtained by collecting the
blackbody radiation of the entrance sapphire window with the use
of an infrared camera. In the spectral regime between 7.5 and 13 μm,
the sapphire window is opaque and, unlike the gaseous ensemble,
its temperature can be measured from its blackbody radiation. The
cooling result, following a 30 s optical cooling period, of a λ = 820
nm laser is presented in Fig. 1.9. The thermal drop induced by the
cooling beam in the gas is depicted in the temperature profile of
the outer sapphire window surface. The measurement indicates a
temperature drop of ΔT = –(0.31 ± 0.03) K near the beam center.
The high thermal conductivity of sapphire (κsapphire = 200 W cm–1
K–1) reduces the here observable cooling. Note that the sapphire
window is in contact with the cell’s metal body, which is heated due
to absorption of scattered fluorescence photons. Nonetheless, the
thermocamera image represents a clear qualitative evidence of laser
cooling based on collisional redistribution.
Thermal deflection measurements are then performed to
accurately determine the temperature change in the gas induced by
the cooling beam. The He–Ne laser beam is moved collinearly with
respect to the cooling beam in steps of 250 μm with the help of a
beam splitter, as shown in Fig. 1.7. The deflection angle is measured

at each position. In this way, the temperature profile induced by
the Ti:sapphire laser beam is resolved in the radial direction. The
probe beam scanning proceeds up to an offset were no deflection is
observed.
Figure 1.9 Average color-coded temperature change after a 30 s optical

cooling period, as measured using an infrared camera directed
at the entrance sapphire window. The blue region in the
center is the sapphire window, cooled by the underlying gas
ensemble. The surrounding metal flange appears red and
yellow due to constant heating, attributed to absorption of
scattered fluorescence photons.
At the position where the two beams are perfectly superimposed,

a symmetric deflection profile is monitored by the position sensitive
photodiode. For any other spatial offset, an asymmetric distribution
of the measured deflection angles is observed. We therefore fit to the
data the heat equation of a temperature gradient as a function of the
thermal diffusivity D, the time t, and the beam waist w0
dT Ê Ê 2r 2 ˆ Ê 2r 2 ˆ ˆ
µ Á exp Á - 2 ˜ - exp Á - 2 ˜˜
(1.18)
dr Ë Ë w0 ¯ Ë 8Dt + w0 ¯ ¯
similarly to previous work [39]. Note that the constant of

proportionality of Eq. 1.18 is determined by the cooling power Pcool
and the thermal conductivity κ. The actual fitting function f(r) is
Ê Ê 2(r - b)2 ˆ Ê 2(r - b)2 ˆ ˆ
dT Ê a ˆ
f (r ) =
= Á exp Á - ˜ - exp Á - ˜˜
dr ÁË r - b ˜¯ Á ÁË w02 ˜¯ ÁË d + w02 ˜¯ ˜
Ë ¯
(1.19)
containing a, b, d, and the beam waist w0 as fitting parameters.
Integration of this devolution leads then to the temperature
expression as a function of the angular deflection φ(r). As several
assumptions were made for deducing this relation, we give an
error on the measured values (position of the probe laser beam,
photodiode deflection signal) but forego an error estimation of the
temperature change. The temperature change calculated out of the
experimental deflection data will in the following be denoted as
ΔTexp. For this calculation, we assume the absorption α to be constant
over the inner cell length.
The results of this measurement are depicted in Fig. 1.10. The
experimentally measured deflection angles (red spheres) are
numerically integrated. This gives a temperature profile as shown by
the solid blue line, with a maximum determined temperature drop of
ΔTexp = −66 K near the cell entrance. For comparison, the deflection
data are fitted assuming a temperature profile from a heat transport
model, giving a temperature profile as shown by the dashed blue
line. Here, the cooling power Pcool and the thermal diffusivity of the
gas, D, were left as free parameters for the fit.
It is worth noting that temperature changes corresponding to a
factor of 2 larger than the one measured in the proof-of-principle
experiment have been observed in subsequent experiments [41].
This is attributed to the higher absorption coefficients and smaller
beam waists used, corresponding to higher excitation intensity.
Further experiments with rubidium as optically active species
include other noble gases as a buffer gas. In the work of Vogl et al.
[47], experiments with helium as a buffer gas are described. The
observed temperature change is smaller in these experiments. This
is in agreement with expectations, since helium has a higher thermal
conductivity than argon.
Figure 1.10 Deflection angle (red spheres) of the He–Ne laser beam as a
function of its lateral offset to the cooling beam for a dense
rubidium-argon mixture (p = 230 bar, T = 620 K). A fit (solid
black line) on the measured data is performed following
Eq. 1.19. The temperature profile near the cell entrance derived
from the deflection data (solid blue line), as well as following a
heat transport model (dashed blue line) are in good agreement.
The maximum determined temperature drop is 66 K.
A rough estimation based on the ideal gas law indicates an

optimum set of parameters for the cooling experiment. When the
temperature T decreases, the density ρ = N/V increases, because the
pressure p in the cooling region is considered constant, in agreement
with the work of Jackson et al. [35].
This affects the fluorescence spectra in a way that the mean
fluorescence starts shifting to the red. When its value becomes equal
to the incident laser frequency, the cooling process is expected to
stop. Therefore, it can be concluded that there is an optimum set
of values for both pressure and wavelength. Experiments examining
the redistributed fluorescence as function of pressure [18] also give
rise to this fact.
We are also currently carrying out further data analysis of
measurements in which a higher temperature change than the
one reported above is observed [41, 48]. However, the model we
presently use for the data analysis neglects possible temperature-
dependent changes of the absorption coefficient α. This is expected
to become an important factor for temperature changes that are of
the same order of magnitude as the absolute temperature of the
ensemble.
Another approach for a determination of the temperature

change can be reached by monitoring the thermal lens induced
by the temperature change using a Shack–Hartmann wavefront
detector [49, 50]. Thus, a new high-pressure cell with four optical
windows will in the future allow for a determination of a 3D
temperature distribution as the cooling region will then be accessible
perpendicularly to the laser beam propagation direction.
1.4.3 Experiments on Potassium–Argon Mixtures

We extended our investigation on the redistribution cooling
technique by performing measurements using potassium atoms as
the optically active species. The fine structure splitting of the excited
state of the potassium D resonances is smaller than that of rubidium
(3 nm compared to 15 nm [51]). This smaller splitting is expected
to better simulate a two-level system, facilitating the interpretation
of the cooling results. For these experiments [29], the high-pressure
cell has to be heated to its temperature limit (T = 720 K) in order
to produce a sufficient particle density. The potential curves for the
potassium-argon quasi-molecule are comparable to those of the
rubidium-argon system [52, 53].
As a first step for the cooling of potassium-argon mixtures,
redistribution of fluorescence has to be proven. Therefore, a setup as
the one described in Section 1.3.2 is used to record the fluorescence
of the dense mixture. Typical potassium spectra recorded at T = 713
K, under 200 bar of argon, for both red- and blue-detuned excitation
wavelengths are depicted in Fig. 1.11. The incident laser power is
PLaser ≈ 600 mW and the beam is focused down to a waist of 2w0
≈ 6 μm. Similarly to the rubidium fluorescence spectrum shown
in Section 1.4.1, clear redistribution toward the center of the D1,2
resonances, accompanied by strong spectral shift and broadening is
observed for both measurements.
More specifically, the resonances are shifted by 3–4 nm to the red
compared to the unperturbed case. Residual scattered light at the
spectral position of the incident light is also present due to imperfect
filtering. On the red wing of the spectrum, the fluorescence level is
much higher than on the higher-energy blue wing. We attribute this
to satellite resonances of the potassium-argon quasi-molecule in

agreement with the work of Kuhn [54] and Kielkopf [55].
Figure 1.11 Fluorescence of a dense potassium–argon mixture at a

pressure of 200 bar and a temperature T = 713 K. The dashed
line denotes the signal recorded for a red-detuned laser at
λ = 780 nm and the solid line for a blue detuning at λ = 765
nm. Vertical lines of identical notation are used to indicate the
spectral position of the incident light in each case. The D1,2
atomic resonances are also depicted (vertical dotted lines).
For different incident wavelengths, red-detuned to the potassium

D resonances, thermal deflection spectroscopy measurements were
conducted, monitoring the deflection profile in steps of 50 μm. A
temperature change of ΔTexp ≈ −120 K was observed for an incident
wavelength of λ ≈ 785 nm and an incident power of PLaser ≈ 1.4 W. This
measurement is depicted in Fig. 1.12. Here, the cooling laser beam is
focused down to a beam diameter of 0.5 mm. The large temperature
change observed in this experiment is mainly attributed to the high
absorption, producing an absorption length of labs = 1.2 mm.
Using the fluorescence spectrum for the incident wavelength
(λ = 785 nm), we can determine the detuning in Eq. 1.1. In this
manner, the cooling power is estimated in the order of 30 mW, which
corresponds to a cooling efficiency of 2%.
Kennard–Stepanov Experiments 23
Figure 1.12 Thermal deflection measurement for a dense potassium-argon

mixture at an initial temperature T = 713 K and a pressure of
200 bar. The deflection angle (black spheres) of the He–Ne laser
beam is shown as a function of its lateral offset to the cooling
beam. A fit (red line) on the measured data is performed
following Eq. 1.19, while integration of the latter yields the
relative temperature change (blue solid line). A temperature
drop by 120 K is observed.
1.5 Kennard–Stepanov Experiments

In the following section we present an alternative approach in
temperature determination of the dense gaseous ensembles, based
purely on spectroscopic data. More specifically, we introduce the
Kennard–Stepanov relation, a Boltzmann-type frequency scaling
connecting the emission and absorption profiles of the investigated
transitions. We begin by offering insight into the theoretical
background of the relation and subsequently present experiments
focused on temperature determination of the gaseous ensembles,
based on the spectral profiles of the rubidium D1,2 atomic transitions.
1.5.1 The Kennard–Stepanov Relation

The investigation of an inherent connection between the absorption
and emission spectral profiles of transitions in strongly interactive
media has been a subject of yearlong research. Following the efforts
of Kennard [56] and Stepanov [57], such a universal relation has been
derived, applying for systems where efficient relaxation processes
lead to thermalization of both the ground and the electronically
excited state manifolds of the addressed transition. The so-called
Kennard–Stepanov relation can be written as follows:
Ê a (w ) ˆ h
ln Á ˜ = w + C(T ) (1.20)
Ë f (w ) ¯ kBT
where α(ω) and f(ω) are the frequency-dependent absorption
and emission line shapes, respectively, T is the so-called spectral
temperature and C(T) is a temperature-dependent constant. Its
derivation is based on the Boltzmann distribution of the occupation
probabilities within both the ground and the electronically excited
state manifolds and can be found elsewhere [58]. In the case of fully
thermalized state manifolds, the equation predicts a linear behavior of
the logarithm versus frequency, with a slope equal to ℏ/kBT. Linearity
deviations are mainly attributed to incomplete thermalization of the
excited state manifold, internal state conversion, and finite quantum
efficiency of the addressed transition. Extraction of the slope using
spectroscopic data of absorption and emission profiles can lead to
temperature determination of the sample. The method has already
been used for studies on dye molecules in liquid solutions [59,
60], solid state inhomogeneous systems [61, 62], and biochemical
photosystems [63], as well as thermalization studies of a photon
gas [64]. Even though an exact temperature extraction has not
been achieved in these experiments, mainly due to inhomogeneous
broadening [65, 66], the thermalization of the relevant states has
been verified through the linearity of the corresponding Kennard–
Stepanov plots. It is however expected that in high-purity gaseous
systems this contribution is strongly suppressed and an exact
temperature determination becomes possible. To this end, the
relation is applied to our dense rubidium–argon gas mixtures.
Before proceeding to the corresponding experiments, we note
the particular conditions under which the validity of the relation
is ensured. Even though the ground-state manifold of a transition
is quite often thermalized, for dilute gases under laser irradiation
this is not usually the case for the electronically excited state.
Most importantly, the thermalization rate is necessary to be much
faster than the radiative lifetime of the excited state, as only in this
case a thermal distribution can arise in the upper electronic state

manifold before radiative decay takes place. A further requirement
for the Kennard–Stepanov relation is the quantum efficiency of the
transition, which needs to be near unity.
The universality of the Kennard–Stepanov relation is proven by
its independence regarding the relaxation processes that lead to
thermalization of the excited state manifold. In solids, interactions
with phonons cause such a thermal distribution, whereas in liquid
or gaseous samples it is frequent collisions that play that role. In
our high-pressure gas ensembles, the number of atomic collisions
within the radiative lifetime (27 ns) of the D1,2 rubidium transition
is of the order of 104 [67]. The latter, combined with the extremely
elastic nature of the collisions between rubidium and noble atoms
[25], leads to efficient thermalization of the electronically excited
quasi-molecular manifold.
1.5.2 Kennard–Stepanov Spectroscopy

Our samples are prepared in the same manner as described in Section
1.3.1. In these spectroscopic measurements a three-window cell is
utilized, while the rubidium optical density is kept low to suppress
reabsorption and preserve quantum efficiency close to unity [25].
We note that here, due to the low optical density requirement, we
aim in extracting the ambient temperature of the ensemble. For this
reason, the Ti:sapphire laser beam has a diameter of 3 mm, while its
power is kept below 1 W, in order to suppress additional heating or
cooling effects. The implications of these requirements are discussed
in Section 1.5.3.
For a complete Kennard–Stepanov study of the rubidium D1,2
transitions [58, 68], pairs of absorption and fluorescence spectra
are recorded in order to produce the relevant plots. We collect
the laser-induced fluorescence in a perpendicular direction with
regard to the laser propagation and analyze it spectrally using a
grating spectrometer. The absorption spectrum, however, cannot
be accurately obtained using regular white light sources, due to
the low optical density of the rubidium vapor. For this reason
here, we employ the method of measuring the fluorescence yield.
More specifically, the absorption spectrum can be reconstructed
by recording the fluorescence power integrated over the emission
spectrum and plotting it versus the incident laser wavelength. In this

manner, we obtain a signal proportional to the absorption coefficient.
The fluorescence yield is additionally tested for different excitation
powers in order to verify that the laser intensity is below saturation.
Furthermore, fluorescence measurements using different excitation
wavelengths allow us to verify that, in the high-pressure regime, the
emission’s spectral profile is largely independent of the excitation
energy. Such a result, also known as Kasha’s rule [69], is depicted in
Fig. 1.13.
Figure 1.13 Fluorescence spectra of the atomic rubidium D1,2 transitions at

573 K, under 190 bar argon buffer gas, collected for different
excitation wavelengths: λ1 = 759 nm (green), λ2 = 785 nm
(blue), λ3 = 799 nm (black), and λ4 = 806 nm (red). The spectra
are normalized due to different efficiency at each excitation
wavelength. Residual scattering of the excitation laser beam is
observed in each case.
A measure of the redistribution that occurs within the excited

state quasi-molecular manifold can be also given through a pressure
dependence of the difference of the observed relative fluorescence
signals at the positions of the rubidium D2 and D1 lines fλ2(D2)/fλ2(D1)
– fλ1(D2)/fλ1(D1). This ratio is averaged over different excitation
wavelength ranges λ1 = 800 – 820 nm and λ 2 = 760 – 780 nm, which
favorably populate the D1 and D2 resonances, respectively. Figure 1.14
shows that for high–buffer gas pressures (above 100 bar), the ratio
difference takes values below 4%, verifying strong redistribution of
the excited state manifold due to the increased number of collisions.
Furthermore, the finer line shape details become wavelength

independent in agreement with previous observations of our group
[17, 28].
Figure 1.14 Pressure dependence of the difference of the relative collected

fluorescence intensity at the position of the rubidium D1 and
D2 resonances (represented by dots), Δ21 = fλ2(D2)/fλ2(D1) –
fλ1(D2)/fλ1(D1), averaged over excitation wavelength ranges λ1=
800–820 nm and λ2= 760–780 nm. An exponential fit is used as
a guide for the eye (solid line).
We collect pairs of absorption and fluorescence spectra for a

wide range of experimental parameters (p, T) of the D1,2 rubidium
transitions. Figure 1.15a depicts such a pair obtained at 420 K,
under 180 bar of argon buffer gas. In both the pressure-broadened
absorption (dashed line) and fluorescence (solid line) spectral
profiles the D1,2 resonances are clearly resolved, overlapping with a
broad spectral wing which extends between 730 nm and 940 nm. A
blue satellite (labelled S) is also visible in both cases. Interestingly,
the spectral profiles exhibit an exactly opposite asymmetry, with the
absorption (fluorescence) having a more pronounced short (long)
wing, similar to the so-called mirror rule observed in dye molecule
solutions [70]. We plot the corresponding Kennard–Stepanov
relation in Fig. 1.15b, where the logarithmic ratio of the absorption
over the fluorescence spectra is depicted versus the optical wave
number. The experimental data (dots) exhibit a clearly linear
behavior, as predicted in Eq. 1.20, for a range that well surpasses
4kBT. Small deviations are observed in the vicinity of D1,2, attributed
to incomplete redistribution, as well as for extreme detunings,
where the collected signal is very weak, yielding a low experimental

signal-to-noise ratio. We perform a linear fit (solid line) and extract
the slope hc/kBT, with T corresponding to the cell’s temperature.
Figure 1.15 (a) Rubidium atomic D1,2 experimental absorption (dashed

line) and fluorescence (solid line) spectral profiles collected
at T = 420 K, under p = 180 bar of argon buffer gas. The
atomic resonances (D1,2) and the blue satellite (S) are noted.
(b) The corresponding Kennard–Stepanov plot depicting the
logarithmic ratio of the experimental spectra (dots). A linear
fit (solid line) with a slope hc/kBT is used as guide for the
eye.
Experiments performed using lower buffer gas pressures (below

100 bar) exhibit also this linear behavior. However, the observed
deviations around the resonances and at the extremes become
significantly higher, emphasizing the role of frequent collisions in
the thermalization process. More importantly, the requirement for a
high collisional rate is compatible with the experimental conditions
under which redistributional laser cooling becomes feasible (see
Section 1.2.2).
More recently, a theoretical and experimental analysis of the
cooling efficiency achieved with the redistribution technique for
media that fulfills the Kennard–Stepanov relation has been carried
out [71]. The calculations well reproduce an observed asymmetry
of the laser cooling and heating efficiencies for red and blue laser
detunings from the pressure broadened rubidium resonance,
respectively, with the efficiency for laser heating (for blue detuning)
being larger than that observed for laser cooling (for red detuning).
This agrees with the intuitive expectations that to heat is easier than
to cool. The results can also be qualitatively understood from the
asymmetry between the absorption and the emission line shapes
for the pressure broadened gas predicted by the Kennard–Stepanov
relation.
1.5.3 Prospects of the Kennard–Stepanov Technique

The experiments described in Section 1.5.2 highlight the potential of
Kennard–Stepanov spectroscopy as a reliable noncontact technique
for temperature determination. However, the main issue that arises
by the use of the atomic D1,2 transitions is that while a high opti-
cal density is necessary for the redistributional cooling experiment,
this is not compatible with the requirements for a simultaneous
Kennard–Stepanov study. More specifically, it is expected that high
optical density conditions favor strong reabsorption effects, limit-
ing the quantum efficiency of the transitions. This obstacle may be
overcome with the use of auxiliary transitions featuring significantly
lower oscillator strengths. Figure 1.16 shows a broadband absorp-
tion spectrum in the dense gaseous ensemble with additional atomic
and molecular transitions [41]. Here, the absorption coefficient of
the D1,2 resonances exhibits strong saturation corresponding to an
optical density sufficiently high to allow for redistributional cooling.
However, the existence of additional atomic and molecular absorp-
tion bands featuring significantly lower values fulfills the prerequi-
site for a simultaneous Kennard–Stepanov investigation. It should be
also pointed out that in such dense ensembles energy pooling effects
occur [30], resulting also in the emission of fluorescence originat-
ing from shorter wavelength transitions when addressing a cooling
laser beam to near the D-line spectral resonances. The carriers that
populate such weaker oscillator strength transitions (as the atomic
5s Æ 6p transition) lie also within the cooling zone, and so it is ex-
pected that their spectral temperature should reflect the reduced
temperature of the cooling experiment. Ongoing experiments have
already produced encouraging results.
Figure 1.16 Absorption coefficient of rubidium vapor at T = 670 K, under

180 bar argon buffer gas, as measured by spectrally analyzing
the transmission of the ensemble using a spectrally broadband
halogen lamp. The atomic and molecular absorption bands are
noted with respect to Ref. [72]. The second principal series
doublet transition, 5s Æ 6p at 420 nm, overlaps with the broad
rubidium dimer X Æ E transition, centered at 430 nm.
1.6 Conclusions
We have reviewed experiments on redistributional laser cooling, a
technique currently applicable to dense alkali–rare gas mixtures.
The achieved cooling powers here are several orders of magnitude
larger than in Doppler cooling experiments for dilute atomic vapors,
and the typically achieved cooling efficiencies are in the order of
several percent.
The alkali transitions are strongly broadened by the high-
pressure gas. The linewidths are in the order of the thermal energy in
frequency units, that is, kBT/h. The cooling principle can be described
in the frame of a quasi-molecular treatment. Elastic collisions of the
optically active species with a surrounding perturber atom enable
optical excitation of the former by laser light, red-detuned to the
atomic resonances. After the collision, the interatomic distance
becomes larger, and the excited state decays under the emission of
a fluorescence photon after its natural lifetime. The photon energy
is near the difference between the ground and the excited state of
an undisturbed alkali atom and is thus higher than the energy of the
References 31
absorbed photon. Energy in the order of the thermal energy kBT is

thus removed from the dense sample, locally cooling the gas.
The cooling results have been presented in the form of thermal
deflection spectroscopy measurements on both rubidium- and
potassium-argon mixtures. The technique allows for an estimation
of the local temperature induced by the cooling beam. Here, the
gas density changes following the temperature change. The former
can be investigated by monitoring the local change of the refractive
index within the so-called cooling zone. Implementation is achieved
by irradiating a second, nonresonant probe laser beam into the high-
pressure cell. The cooled gas acts like a graded index lens, causing
the probe beam to deflect. Position dependent measurements of the
deflection angle, varying the distance between the cooling and probe
beams, allow for a mapping of the deflection profile, from which the
temperature change can be deduced.
We have further discussed an alternative approach of
temperature determination utilizing purely spectroscopic data.
To this end, we have presented the Kennard–Stepanov relation,
connecting absorption and emission profiles, and applied it in our
dense gaseous ensembles. Our investigation shows that the relation
is well fulfilled allowing for an accurate temperature determination
of the ensemble’s ambient temperature. Further studies on auxiliary
atomic and molecular transitions are currently in progress.
Future experimental aims include the study of homogeneous
nucleation in saturated vapor, as well as the application of this
very technique to molecular–rare gas ensembles. Toward this goal,
ongoing preparatory experiments seem encouraging [48, 73].
References
1. Pringsheim, P. (1929). Zwei Bemerkungen über den Unterschied von

Lumineszenz- und Temperaturstrahlung, Z. Phys., 57(11–12), pp. 739–
746.
2. Maiman, T. H. (1960). Stimulated optical radiation in ruby, Nature,
187(4736), pp. 493–494.
3. Hänsch, T. W., Schawlow, A. L. (1975). Cooling of gases by laser
radiation, Opt. Commun., 13(1), pp. 68–69.
4. Wineland, D. J., Itano, W. M. (1979). Laser cooling of atoms, Phys. Rev.
A,, 20(4), 1521–1540.
5. Adams, C. S., Riis, E. (1997). Laser cooling and trapping of neutral

atoms, Prog. Quant. Electron., 21(1), pp. 1–79.
6. Lett, P. D., Watts, R. N., Westbrook, C. I., et al. (1988). Observation of
atoms laser cooled below the doppler limit, Phys. Rev. Lett., 61(2), pp.
169–172.
7. Anderson, M. H., Ensher, J. R., Matthews, M. R., et al. (1995). Observation
of Bose-Einstein condensation in a dilute atomic vapor, Science,
269(5221), pp. 198–201.
8. Davis, K. B., Mewes, M., Andrews, M. R., et al. (1995). Bose-Einstein
condensation in a gas of sodium atoms, Phys. Rev. Lett., 75(22), pp.
3969–3973.
9. Shuman, E. S., Barry, J. F., Demille, D. (2010). Laser cooling of a diatomic
molecule, Nature, 467(7317), pp. 820–823.
10. Kastler, A. (1950). Quelques suggestions concernant la production
optique et la détection optique d’une inégalité de population
des niveaux de quantifigation spatiale des atomes. Application à
l’expérience de Stern et Gerlach et à la résonance magnétique, J. Phys.
Radium, 11(6), pp. 255–265.
11. Djeu, N., Whitney, W. T. (1981). Laser cooling by spontaneous anti-
Stokes scattering, Phys. Rev. Lett., 46(4), pp. 236–239.
12. Epstein, R. I., Buchwald, M. I., Edwards, B. C., et al. (1995). Observation
of laser-induced fluorescent cooling of a solid, Nature, 377(6549), pp.
500–503.
13. Seletskiy, D. V., Melgaard, S. D., Bigotta, S., et al. (2010). Laser cooling of
solids to cryogenic temperatures, Nat. Photonics, 4(3), pp. 161–164.
14. Yeh, S., Berman, P. R. (1979). Theory of collisionally aided radiative
excitation in three-level systems: new interference effect, Phys. Rev.
Lett., 43(12), pp. 848–851.
15. Berman, P. R., Stenholm, S. (1978). Heating or cooling collisionally
aided fluorescence, Opt. Commun., 24(2), pp. 155–157.
16. Giacobino, E., Tawil, M., Berman, P. R., et al. (1983). Production of “hot”
excited-state atoms in collisionally aided radiative transitions, Phys.
Rev. A, 28(4), pp. 2555–2557.
17. Vogl, U., Weitz, M. (2009). Laser cooling by collisional redistribution of
radiation, Nature, 461(7260), pp. 70–73.
18. Carlsten, J. L., Szöke, A., Raymer, M. G. (1977). Collisional redistribution
and saturation of near-resonance scattered light, Phys. Rev. A, 15(3), pp.
1029–1045.
References 33
19. Burnett, K. (1985). Collisional redistribution of radiation, Phys. Rep.,

118(6), pp. 339–401.
20. Carlsten, J. L., Szöke, A. (1976). Spectral resolution of near-resonant
Rayleigh scattering and collision-induced resonance fluorescence,
Phys. Rev. Lett., 36(12), pp. 667–671.
21. Huber, D. L. (1969). Resonant scattering of monochromatic light in
gases, Phys. Rev., 178(1), pp. 93–102.
22. Omont, A., Smith, E. W., Cooper, J. (1972). Redistribution of resonance
radiation. I. The effect of collisions, Astrophys. J., 175, p. 185.
23. Vogl, U. (2010). Kollektive Effekte und stoßinduzierte Redistribution-
skühlung in dichten atomaren Gasen, PhD thesis.
24. Dhiflaoui, J., Berriche, H., Heaven, M. C., et al. (2011). Electronic
Absorption Spectra of the RbAr van der Waals Complex. In Proceedings
of the fifth Saudi Physical Society Conference (SPS5), AIP, pp. 234–241.
25. Speller, E., Staudenmayer, B., Kempter, V. (1979). Quenching cross
sections for alkali-inert gas collisions, Z. Phys. A, 291(4), pp. 311–318.
26. Mungan, C. E., Gosnell, T. R. (1999). Laser Cooling of Solids. Vol. 40,
Elsevier, pp. 161–228.
27. Rayner, A., Heckenberg, N. R., Rubinsztein-Dunlop, H. (2003).
Condensed-phase optical refrigeration, J. Opt. Soc. Am. B, 20(5), p.
1037.
28. Vogl, U., Saβ, A., Haβelmann, S., et al. (2011). Collisional redistribution
laser cooling of a high-pressure atomic gas, J. Mod. Opt., 58(15), pp.
1300–1309.
29. Saß, A., Vogl, U., Weitz, M. (2011). Laser cooling of a potassium–argon
gas mixture using collisional redistribution of radiation, Appl. Phys. B,
102(3), pp. 503–507.
30. Allegrini, M., Gabbanini, C., Moi, L. (1985). Energy pooling processes
in laser-excited alkali vapors: an update on experiments, J. Phys.
Colloques, 46(C1), pp. C1-61–C1-73.
31. Barbier, L., Cheret, M. (1983). Energy pooling process in rubidium
vapour, J. Phys. B, 16(17), pp. 3213–3228.
32. Bondi, A. (1964). van der Waals Volumes and Radii, J. Phys. Chem.,
68(3), pp. 441–451.
33. Bridgman, P. W. (1946). Recent work in the field of high pressures, Rev.
Mod. Phys., 18(1), pp. 1–93.
34. Whinnery, J. R. (1974). Laser measurement of optical absorption in
liquids, Acc. Chem. Res., 7(7), pp. 225–231.
35. Jackson, W. B., Amer, N. M., Boccara, A. C., et al. (1981). Photothermal
deflection spectroscopy and detection, Appl. Opt., 20(8), pp. 1333–
1344.
36. Spear, J. D., Russo, R. E., Silva, R. J. (1990). Collinear photothermal
deflection spectroscopy with light-scattering samples, Appl. Opt.,
29(28), pp. 4225–4234.
37. Casperson, L. W. (1973). Gaussian light beams in inhomogeneous
media, Appl. Opt., 12(10), pp. 2434–2441.
38. Born, M., Wolf, E. (1999). Principles of optics: electromagnetic theory
of propagation, interference and diffraction of light/Max Born and
Emil Wolf with contributions by A. B. Bhatia, et al., 7th (expanded) ed.
Cambridge University Press, Cambridge.
39. Gordon, J. P., Leite, R. C. C., Moore, R. S., et al. (1965). Long-transient
effects in lasers with inserted liquid samples, J. Appl. Phys., 36(1), p. 3.
40. Zhu, Q., Pan, B., Chen, L., et al. (2010). Analysis of temperature
distributions in diode-pumped alkali vapor lasers, Opt. Commun.,
283(11), pp. 2406–2410.
41. Saß, A. (2014). Experimente zur stoßinduzierten Redistributions-
laserkühlung dichter atomarer Alkali-Edelgas-Mischungen und
Simulationen der zu erwartenden Temperaturverteilung, PhD thesis.
42. Shang-Yi, C. (1940). Broadening, asymmetry and shift of rubidium
resonance lines under different pressures of helium and argon up to
100 atmospheres, Phys. Rev., 58(12), pp. 1051–1058.
43. Daniel A. Steck, “Rubidium 85 D Line Data,” available online at http://
steck.us/alkalidata (revision 2.1.6, 20 September 2013).
44. Drummond, D. L. (1974). Potentials and continuum spectra of Rb-
noble gas molecules, J. Chem. Phys., 60(9), p. 3426.
45. Alcock, C. B., Itkin, V. P., Horrigan, M. K. (1984). Vapour pressure
equations for the metallic elements: 298–2500K, Can. Metall. Quart.,
23(3), pp. 309–313.
46. Mayer, H. (1931). Über eine neue Methode zur Messung kleinster
Dampfdrucke. Die Dampfdrucke von Quecksilber und Kalium, Z. Phys.,
67(3–4), pp. 240–263.
47. Vogl, U., Saß, A., Weitz, M. (2012). Laser cooling of dense rubidium-
noble gas mixtures via collisional redistribution of radiation. In
Epstein, R. I., Sheik-Bahae, M. (eds.) SPIE OPTO. SPIE, pp. 827508-1–
827508-8.
48. Saß, A., Forge, R., Christopoulos, S., et al. (2014). Laser cooling of
dense atomic gases by collisional redistribution of radiation and
References 35
spectroscopy of molecular dimers in a dense buffer gas environment.

In Epstein, R. I., Seletskiy, D. V., Sheik-Bahae, M. (eds.) SPIE OPTO. SPIE,
p. 90000A.
49. Hartmann, J. (1900). Bemerkungen über den Bau und die Justierung
von Spektrographen, Z. Instrumentenkd, 20, p. 47.
50. Shack, R. V., Platt, B. C. (1971). Production and use of a lenticular
Hartmann screen. In Chairman, D. S. (ed.) Spring Meeting of the Optical
Society of America. Optical Society of America.
51. Tiecke, T. (ed.) (2010). Properties of Potassium. PhD Thesis, University
of Amsterdam, The Netherlands.
52. Rhouma, M. B. El Hadj, Berriche, H., Lakhdar, Z. B., et al. (2002). One-
electron pseudopotential calculations of excited states of LiAr, NaAr,
and KAr, J. Chem. Phys., 116(5), p. 1839.
53. Figl, C. (2004). Optical collisions in crossed beams and Bose-Einstein
condensation in a microtrap, PhD thesis.
54. Kuhn, H., Oldenberg, O. (1932). Evidence of space quantization of
atoms upon impact, Phys. Rev., 41(1), pp. 72–78.
55. Kielkopf, J. F. (1968). Semiclassical theory of satellite bands produced
in the spectra of alkali metals by interaction with foreign gases, J. Chem.
Phys., 48(12), p. 5570.
56. Kennard, E. H. (1918). On the thermodynamics of fluorescence, Phys.
Rev., 11(1), pp. 29–38.
57. Stepanov, B. I. (1957). Universal relation between the absorption
spectra and luminescence spectra of complex molecules, Dokl. Akad.
Nauk SSSR, 112, p. 839.
58. Moroshkin, P., Weller, L., Saß, A., et al. (2014). Kennard-Stepanov
relation connecting absorption and emission spectra in an atomic gas,
Phys. Rev. Lett., 113(6), p. 063002.
59. Sawicki, D. A., Knox, R. S. (1996). Universal relationship between
optical emission and absorption of complex systems: an alternative
approach, Phys. Rev. A, 54(6), pp. 4837–4841.
60. Dobek, K. (2011). The influence of temperature on Coumarin 153
fluorescence kinetics, J. Fluoresc., 21(4), pp. 1547–1557.
61. Band, Y. B., Heller, D. F. (1988). Relationships between the absorption
and emission of light in multilevel systems, Phys. Rev. A, 38(4), pp.
1885–1895.
62. Ihara, T., Maruyama, S., Yoshita, M., et al. (2009). Thermal-equilibrium
relation between the optical emission and absorption spectra of a
doped semiconductor quantum well, Phys. Rev. B, 80(3), p. 033307.
63. Zucchelli, G., Garlaschi, F. M., Croce, R., et al. (1995). A Stepanov
relation analysis of steady-state absorption and fluorescence spectra
in the isolated D1/D2/cytochrome b-559 complex, Biochim. Biophys.
Acta, 1229(1), pp. 59–63.
64. Klaers, J., Schmitt, J., Vewinger, F., et al. (2010). Bose-Einstein
condensation of photons in an optical microcavity, Nature,
468(7323), pp. 545–548.
65. van Metter, R., Knox, R. (1976). On the relation between absorption
and emission spectra of molecules in solution, Chem. Phys., 12(3), pp.
333–340.
66. Kawski, A., Bojarski, P., Kukliński, B. (2000). The local temperature
dependence of fluorescent centres in PVA films on the excitation
wavenumber, Z. Naturforsch. A, 55(6–7), pp. 653–655.
67. Ottinger, C., Scheps, R., York, G. W., et al. (1975). Broadening of the Rb
resonance lines by the noble gases, Phys. Rev. A, 11(6), pp. 1815–1828.
68. Christopoulos, S., Saß, A., Moroshkin, P., et al. (2015). A Kennard-
Stepanov relation study on redistributional laser cooling in dense
gaseous ensembles. In Epstein, R. I., Seletskiy, D. V., Sheik-Bahae, M.
(eds.) SPIE OPTO. SPIE, p. 93800M.
69. Kasha, M. (1950). Characterization of electronic transitions in complex
molecules, Discuss. Faraday Soc., 9, p. 14.
70. Lakowicz, J. R. (2006). Principles of Fluorescence Spectroscopy. Springer
US, Boston, MA.
71. Gelbwaser-Klimovsky, D., Szczygielski, K., Vogl, U., et al. (2015). Laser-
induced cooling of broadband heat reservoirs, Phys. Rev. A, 91(2), p.
023431.
72. Vdović, S., Sarkisyan, D., Pichler, G. (2006). Absorption spectrum
of rubidium and cesium dimers by compact computer operated
spectrometer, Opt. Commun., 268(1), pp. 58–63.
73. Simon, P., Moroshkin, P., Weller, L., et al. (2013). Towards redistribution
laser cooling of molecular gases: production of candidate molecules
SrH by laser ablation. In Epstein, R. I., Seletskiy, D. V., Sheik-Bahae, M.
(eds.) SPIE OPTO. SPIE, p. 86380C.
Chapter 2
Laser Cooling in Rare Earth–Doped

Glasses and Crystals
Galina Nemova
Department of Engineering Physics, Polytechnique Montréal, C.P. 6079,
Succursale Centre-ville, Montréal (Québec) Canada H3C 3A7
galina.nemova@polymtl.ca
2.1 Introduction
Laser cooling is a physical process by which a system loses its
thermal energy as a result of interaction with laser light. The term
“laser cooling” is most often used to describe cooling and trapping
of atoms and ions to extremely low temperatures down to the nano-
Kelvin regime. The energy of molecules in gases, especially at low
temperatures, is more contained in translational degrees of freedom
then in vibrations. Laser cooling in such a system is based on the
Doppler’s effect. The basic idea is to red-shift the frequency of the
light to just below the resonance frequency of the atoms. In that case
the atoms moving toward the laser are blue-shifted into resonance.
They more strongly absorb photons then the atoms, which
move away from the laser and are red-shifted further away from

www.panstanford.com
38 Laser Cooling in Rare Earth–Doped Glasses and Crystals
resonance [1]. Adsorbing a photon an atom moving toward the laser

loses momentum equal to the photon moment. The excited atom
spontaneously relaxes with photon emission in a random direction,
on average, without momentum change. The three-dimensional
orthogonal setup with counterpropagating laser beams creates an
optical molasses, which can be cooled down to the Doppler cooling
limit. This area of science has progressed immensely. In 1997, Chu,
Cohen-Tannoudji, and Phillips were awarded the Nobel Prize in
Physics for developing the methods to cool and trap atoms with light
[2]. While the Doppler cooling technique works for atoms in dilute
gases it is not suitable for cooling of macroscopic objects like solids.
Indeed, in solids translational motion is absent and the thermal
energy is contained in vibrational modes known as phonons. The
vast majority of luminescent materials emit photons with lower
energy than absorbed ones. The associated energy difference is
the Stokes shift, named after the physicist George G. Stokes. The
Stokes shift is based on electron–phonon coupling, which allows
pump light to interact with the vibrational modes of solids. This
process is accompanied by heat generation in the system. If the
emitted photon has more energy than the absorbed one the energy
difference between the emitted and absorbed photons is called
the anti-Stokes shift and the emitted photons are termed “anti-
Stokes fluorescence.” This effect is based on dissipation of thermal
phonons in a crystal lattice. In 1928, Lenard showed that anti-Stokes
fluorescence does not violate the second law of thermodynamics
[3]. In 1929, Pringsheim proposed to use anti-Stokes fluorescence
to cool sodium vapor [4]. He proposed to pump sodium vapor
in “Dewar 1” with a filtered sodium vapor lamp in order to excite
electrons from the 2S1/2 ground state to the first 2P1/2 excited state.
After thermalization, which is based on inelastic collisions in the gas,
to the higher 2P3/2 excited state the electrons relax to the ground
state with resonant (2P1/2 Æ 2S1/2) and anti-Stokes (2P3/2 Æ 2S1/2)
fluorescence. This fluorescence has to be quenched by nonradiative
relaxation with heat generation in “Dewar 2” filled with the mixture
of sodium vapor and nitrogen. In this experiment “Dewar 1” is cooled
with light. Pringsheim assumed that the emitted photons scatter
in all direction and this process is irreversible in principle. Vavilov
believed that optical cooling by anti-Stokes fluorescence contradicts
the second law of thermodynamics [5, 6]. He argued that the cycle,
which includes excitation and fluorescence, is a reversible one and
Introduction 39
the scattered light can be with sufficient approximation once again

converted into a directed beam. Vavilov referred to a Lomonosov’s
paper published in 1743, in which Lomonosov shows the manner
in which the full light flux from a point source can be directed as a
parallel beam [7]. In 1946, Landau concluded this controversy [8].
He considered the entropy of the radiation field based on Bose–
Einstein statistics applied to a “photon gas.” Landau has shown
that the entropy of the radiation is a function of the angle of the
propagating radiation and its frequency bandwidth. Consequently,
laser light has low entropy as compared to higher-entropy anti-
Stokes fluorescence. The laser cooling process based on anti-Stokes
fluorescence does not violate the second law of thermodynamics.
In 1950, Kastler [9] and in 1961 Yatsiv [10] suggested the use of
rare earth (RE)-doped solids for laser cooling. The main advantage
of RE ions is the optically active 4f electrons shielded by the filled
5s and 5p outer shells, which limit the interaction of a RE ion with
the surrounding lattice and suppress nonradiative decay, especially
in low-phonon hosts. In 1968, Kushida and Geusic tried to cool a
neodymium-doped yttrium aluminium garnet (Nd3+:YAG) sample
pumped at 1064 nm and placed inside the cavity of the Nd3+:YAG
laser [11]. They observed a reduction of heating. In 1981, Djeu and
Whitney reached net laser cooling with anti-Stokes fluorescence in
low-pressure CO2 gas with the temperature drop of ~1 K from 600
K using a CO2 laser as a pump source [12]. The net radiation cooling
process based on anti-Stokes fluorescence in solids was observed
for the first time only in 1995 by Epstein’s research team in Yb3+-
doped ZBLANP glass [13]. A 1 wt% Yb3+-doped sample of ZBLANP
of volume 2.5 × 2.5 × 6.9 mm3 was placed in vacuum, pumped in the
870–1050 nm wavelength range and cooled by 0.3 K below room
temperature when the pump wavelength was greater the mean
fluorescence wavelength. In 2013, laser cooling down to 119 K that
is below National Institute of Standards and Technology (NIST)
cryogenic temperature (123 K) was achieved in a ytterbium-doped
yttrium lithium fluoride (Yb3+:YLF) sample [14]. In the same 2013 a
Yb3+:YAG sample of volume 1 × 1 × 10 mm3 was laser-cooled for the
first time in air [15]. Today’s record temperature in vacuum is 91 K.
It has been achieved in a Yb3+:YLF sample [16, 17].
In RE-doped glasses and crystals the laser cooling process takes
place in localized active RE ions; the ground and excited levels of
RE ions as well as phonons of the host are involved in the cooling
cycle. Semiconductors are an alternative class of materials suitable

for laser cooling with anti-Stokes fluorescence. In semiconductors,
the cooling cycle involves transitions between the valence and
conduction bands of direct-band-gap semiconductors. Unfortunately,
the purity of semiconductor samples, surface recombination, high
refractive index of semiconductors in comparison with glasses
and crystals are the obstacles on the way of laser cooling in bulk
semiconductors [18]. In 2013, Zhang et al. observed for the first
time laser cooling of CdS nanobelts with a temperature drop of 40 K
from room temperature [19]. Laser cooling in bulk semiconductors
has not been achieved yet. The progress in laser cooling of solids is
summarized in Table 2.1.
Table 2.1 The history of laser cooling with anti-Stokes fluorescence
Year Achievement References

1920s ASF was observed.
1928 ASF does not violate SLT. [3]
1929 Cooling with ASF was proposed. [4]
1946 The entropy of a “photon gas” was estimated. [8]
1950 RE-doped solids were proposed for cooling. [9]
1968 Nd3+:YAG—reduced heating with ASF. [11]
1981 CO2—net laser cooling with ASF (ΔT = 1 K). [12]
1995 Yb3+:ZBLANP—net laser cooling with ASF [13]
(ΔT = 0.2 K).
2013 Yb3+:YLF–cooling below NIST cryogenic [14]
temperature 123 K.
2013 Yb3+:YAG—net laser cooling in air (ΔT = 8.8 K). [15]
2013 CdS nanobelts—net laser cooling (ΔT = 40 K). [19]
2015 Yb3+:YLF—today’s record temperature (T = 91 K). [17]
ASF, anti-Stokes fluorescence; SLT, the second law of thermodynamics; ΔT, the
temperature drop.
2.2 Thermodynamics of Laser Cooling of Solids

For the first time the second law of thermodynamics was applied to
luminescence processes by Landau in 1946 [8]. In 1960, Weinstein
Thermodynamics of Laser Cooling of Solids 41
estimated the limiting efficiency of electroluminescent processes

[20]. In 1969, Chukova considered luminescence in general way
and obtained the relation for the upper limit of efficiency [21, 22].
Chukova has shown that the limiting efficiency of photoluminescence
cannot be larger than the thermodynamic limit on the efficiency of
electroluminescence [23, 24].
Let us consider a system consisting of the pump radiation, a
luminescent body, which is in thermal equilibrium with an ambient,
and the fluorescence radiation. In the steady-state condition, the
energy conservation requires that
E + Q - E = 0 (2.1)
p f
where E p is the pump laser radiation rate, Q is the heat flow power
delivered to the body, and E f is the fluorescence radiation rate. The
points over the symbols denote the derivative d/dt, where t is time.
The second law of thermodynamics requires that the entropy of an
isolated system does not decrease. It gives rise to the inequality
Q
Sf - - Sp ≥ 0 (2.2)
T
where Sp is the entropy of the pump source, Sf is the entropy of
fluorescence radiation, and T is ambient temperature. The entropy of
radiation (“photon gas”) is determined by the well-known formulae
of Bose statistic.
ln 2
S =kB ÚÚ c 3
[(1 + nn )ln(1 + nn ) - nn ln nn ]dn dW (2.3)
where nν is the mean photon occupation number per unit volume

of the k-space, where k is the wave vector. In the case of polarized
radiation l = 1, l = 2 specifies unpolarized radiation.
The entropy flux density of the radiation in a frequency interval
(ν, ν + dν), which passes the unit area of a surface per unit time into
solid angle dΩ in a direction under an angle θ with the normal to the
surface (Fig. 2.1) is
ln 2
S = k B ÚÚ c 2
[(1 + nn )ln(1 + nn ) - nn ln nn ]cos q dn dW (2.4)
Figure 2.1 Entropy flux density of the radiation emitted from the surface
of a sample into a solid angle dΩ. The polar angle θ is measured
relative to the surface normal direction n.
The factor cos θ projects the unit area of the surface to be normal
to radiation. The energy flux density of the same radiation can be
described as
ln 2
E = ÚÚ c 2
hn nn cos q dn dW (2.5)
The effective flux temperature, which is the mathematical

substitution, can be introduced as the relation between the energy
flux density and entropy flow density
E (2.6)
T eff =
S
2.2.1 Entropy for Near-Monochromatic Radiation Flow

Suppose that the radiation flow is spread in the frequency interval
∆ν around a central frequency ν0 (∆ν << ν0) and propagates into the
angle ∆Ω. In this case Eqs. 2.4 and 2.5 can be written as
ln 2
S = k B 2 [(1 + nn )ln(1 + nn ) - nn ln nn ]cos qDWDn (2.7)
c
ln 2
E = 2 hn nn cos qDWDn (2.8)
c
respectively. Let us suppose that ether ∆ν or ∆Ω go to zero. Since
the energy flow density is nonzero the relation nν∆Ω∆ν in Eq. 2.8
must be finite, that is, nν tends to infinity (nν Æ ∞). Consequently,
Fundamentals of Laser Cooling in Rare Earth–Doped Solids 43
considering Eq. 2.7 for the entropy flow density one can conclude
that for monochromatic or unidirectional radiation fluxes the
entropy flow density vanishes. For the first time that was proved by
Landau in 1945. Using Eq. 2.6 one can conclude that the effective flux
temperature for monochromatic or unidirectional radiation fluxes
approaches infinity (Teff Æ ∞).
2.2.2 Conversion Efficiency

The maximum conversion efficiency, η, of the pump energy flux
density, E p , into fluorescence energy flux density, E f , can be
estimated with Eqs. 2.1, 2.2, and 2.6.
E f Tfeff Ê T ˆ
h= £ Á 1 - eff ˜ (2.9)
E p (Tfeff - T ) ÁË Tp ˜¯
where Tfeff and Tpeff are the effective flux temperature of the fluo-
rescence and the pump radiation, respectively. In the case of near-
monochromatic pump radiation the effective flux temperature
Tpeff Æ ∞ and, as one can see in Eq. 2.9, the maximum conversion ef-
ficiency depends only on the fluorescence effective flux temperature
Tfeff . This result for the first time was obtained by Weinstein in 1960
for electroluminescence [20].
2.3 Fundamentals of Laser Cooling in Rare

Earth–Doped Solids
2.3.1 The Two-Level Model of Laser Cooling in Rare

Earth–Doped Hosts
RE ions (lanthanides) have a long history of optical applications.
REs have unique characteristics that distinguish them from other
optical ions. RE ions absorb and emit over narrow frequency ranges.
The frequencies of absorption and emission transitions are almost
insensitive to host materials. The lifetimes of metastable levels are
long (μs to ms). These optical properties of RE ions are dictated
by their unique electronic structure. All the RE ions have the same
outer electronic structure 5s25p66s2, which are filled shells. The

optical properties of the RE ions are connected with the electrons
occupying the inner 4f shielded quite effectively orbital. As a result,
the RE ions maintain much of their “free ion” properties when are
doped into hosts than other ions. For example, the absorption and
emission spectra of the REs are less dependent on external electric
field than the absorption and emission spectra of the transition
elements, which do not experience electronic shielding. While the
crystal field interactions are weak in RE ions, they are suitable for
laser cooling. The idea to use the RE ions for cooling was proposed
by Kastler in 1950 [9].
Let us consider the laser cooling process with ytterbium ions,
Yb3+, in the two-level system. Laser cooling in the four-level system
has been considered in Refs. [25, 26]. Ionization of the REs usually
results in trivalent states. As an example the energy-level diagram of
Yb3+ ions doped in the YAG host is presented in Fig. 2.2. The energy
gap between the 2F7/2 and 2F5/2 manifolds is ~10,000 cm–1.
Figure 2.2 Energy levels and major transitions of Yb3+ in YAG.
When a free ion is put into a host (in general substitutionally)

it is surrounded by a set of nearest-neighbor ligands in some
configuration. The host ligands and the dopant ions interact
with each other through an electrostatic Coulomb interaction.
A ligand field induces a Stark effect, which results in the splitting
of the energy levels. The thermalization process, which is based
on phonon absorption and creation, occurs within each manifold
(Fig. 2.2). It provides Boltzmann’s distribution of electrons in Stark
split sublevels. The condition of the thermal equilibrium can be

presented as
- +
ANR Nnm = ANR Nn ,m-1 (2.12)
Here Nnm and Nn,m–1 are the population of each sublevel; n = 0, 1 is
the number of the manifold: ground (0) and excited (1); m is the
sublevel number; m = 1, 2, 3, and 4 for the ground manifold; and m =
- +
1, 2, and 3 for the excited manifold. ANR and ANR are the de-excitation
and excitation nonradiative rates, respectively (Fig. 2.2). If ΔEn,m =
En,m – En,m–1 is the energy difference between sublevels (n, m) and (n,
m–1) in the n-manifold, one can obtain the relation
Nnm A+ Ê DEn ,m ˆ
= NR = exp Á - (2.13)
-
Nn ,m-1 ANR Ë kBT ˜¯
where kB is Boltzmann’s constant. Of course the system has to be

free from a “phonon bottleneck,” that is, phonons with the energy,
which permits to bridge sublevels of manifolds have to be generated
in the system. The population of each sublevel can be described
with the Boltzmann occupation factor, fnm, which is a function of the
temperature, T. The Boltzmann occupation factors of the lower and
the upper manifolds can be calculated with relations
Ê E - E0i ˆ
exp Á 01
Ë kBT ˜¯
f0i = (2.14)
4
Ê E - E0i ˆ
Â
i =1
exp Á 01
Ë k T ˜¯
B
and
Ê E11 - E 1 j ˆ
exp Á ˜
Ë kBT ¯
f1 j = (2.15)
3
Ê E11 - E1 j ˆ
Â
j =1
exp Á
Ë kBT ¯
˜
respectively. Here index 0i (i is integer from 1 to 4) refers to the

sublevels of the lower 2F7/2 manifold, and index 1j (j is integer from 1
to 3) refers to the sublevels of the upper 2F5/2 manifold. E0i and E1j are
the energies of the sublevels of the lower 2F7/2 and the upper 2F5/2
manifolds, respectively. The occupation factors obey the equations
4 3
Â
i =1
f0i = Âf
j =1
1j = 1 (2.16)
Each of the Stark sublevels of the 2F5/2 manifold can be

characterized with the branching ratios, b01ij , describing the fraction
of particles of the sublevel of 1j (j = 1, 2, 3) of the upper 2F5/2 manifold,
which decays down to the sublevel 0i (i = 1, 2, 3, 4) of the lower
2F tr
7/2 manifold. 1 j are the lifetimes of the sublevels of the F5/2
2
manifold. The difficulty of spectrally resolving and characterizing

the individual Stark sublevels and transitions result in uncertainty
about these parameters. As a consequence the mean fluorescence
frequency of the 2F5/2 manifold is more useful parameter, which can
be described with the relation
3 4
f1 j
ÂÂ t
j =1 i =1
r
1j
b01ijn01ij
nf = (2.17)
1
tr
where
3
f1 j
Ât
1
= (2.18)
tr r
j =1 1 j
Here τr is the average radiative lifetime of the upper 2F5/2

manifold commonly known simply as the radiative lifetime of the
excited manifold. The mean fluorescence frequency, νf, does not
depend on the pump frequency, νp, (Vavilov’s law) and also can be
estimated with the integral relation
nf =
Ú F (n )ndn (2.19)
Ú F (n )dn
where Φ(ν) is the photon flux density. The population of the lower
2F and the upper 2F5/2 manifolds can be calculated with the
7/2
relations
4 4
N0 = ÂN = Â f
i =1
0i
i =1
0i N0 , and (2.20)
3 3
N1 = Â
j =1
N1 j = Âf
j =1
1 j N1 (2.21)
respectively. The laser cooling process in a two-level system can be

simulated with the well-known rate equations [27, 28]
Ï dN1 Ip N N
Ô = [N0s a (n p ) - N1s e (n p )] - 1 - 1
Ì dt hn p t r t nr (2.22)
Ô
Ó NT = N0 + N1
where Ip is the pump intensity at the frequency νp. σa,e(νp) is the
absorption (a) and emission (e) cross sections at the pump frequency
νp. τnr is the nonradiative lifetime of the upper 2F5/2 manifold. NT is the
RE ions density in the sample. The cooling power density generated
in the sample, Pcool, can be estimated as a difference between the
absorbed and radiated power densities.
hn f hn
Pcool = - Ip [N0s a (n p ) - N1s e (n p )] + N1 - N1 f (2.23a)
tr t nr
For a system in the steady-state condition, dN0/dt = dN1/dt = 0,
one can calculate the cooling power using rate equations (Eq. 2.22)
(hnf - hn p ) Iss a (n p )NT
Pcool = (2.23b)
hn p Ê s e (n p ) Is ˆ
Á1 + + ˜
Ë s a (n p ) Ip ¯
where n f ª hqn f is the effective mean fluorescence frequency.

Is = hvp/[τfσa(vp)] is the saturation intensity, where τf = ηqτr is the
intrinsic fluorescence lifetime of the excited manifold. Here
Wr
hq = (2.24)
Wr + Wnr
is the emission quantum efficiency or internal quantum efficiency,
which describes the “competition” between the radiative, Wr = 1/τr,
and nonradiative, Wnr = 1/τnr, decay rates of the upper 2F5/2
manifold. In real samples the host material and impurities can
absorb the pump photons, too. Denote αb an average background
absorption coefficient. For a real system one has to replace the
emission quantum efficiency, ηq, with the relative emission quantum

efficiency
ar
hq = hq = hqhabs (2.25)
(a r + a b )
where αr ≈ σa(νp)NT is the resonant absorption coefficient at the
relatively low pump power intensity. ηabs is the absorption efficiency,
which can be defined as the fraction of pump laser photons engaged
in cooling. The cooling efficiency can be estimated as the relation of
the cooling power density and absorbed power density
Pcool (hnf - hn p ) Is
hcool ( Ip ) = = (2.26)
Pabs hn p Ê s e (n p ) ˆ
Á Is + I ˜
Ë s a (n p ) p ¯
As one can see in Eq. 2.26 the cooling efficiency is a function

of the pump intensity Ip. As the pump intensity increases the
intensity of the stimulated emission, which does not result in the
cooling process, increases. For relatively weak pump intensities, Ip,
most of the absorbed pump radiation leaves the sample with the
spontaneous emission at the mean fluorescence frequency, νf, and
the cooling efficiency can be estimated as the difference between the
energy of the mean fluorescence and pump photon normalized by
the energy of the pump photon
hnf - hn p lp - lf
hcool ª = (2.27)
hn p fl
here λp = c/vp and lf = c nf are the pump wavelength and the
mean fluorescence wavelength, respectively. c is the speed of light
in vacuum. As one can see in Eq. 2.26 if the pump power intensity
reaches saturation IpSat = hn p (t f s e (n p )) the cooling efficiency, ηcool,
decreases in half of the cooling efficiency corresponding to the
weak pump intensity, which is described with Eq. 2.27. The cooling
efficiency described with Eq. 2.26 takes place in an ideal unrealistic
sample where reabsorption and the total internal reflection (TIR)
of radiated photons at the boundaries of the sample are not taken
into account. In actual samples used in experiments the process of
reabsorption, TIR as well as the cooperative effects, such as energy
migration and cooperative luminescence, which become essential at

high ion concentrations, influence the fluorescence lifetime of the
excited manifold, τf, and the cooling efficiency of the system.
Nonradiative decay of the excited level in RE ions with many
excited manifolds such as Tm3+, Er3+ can include the multiphonon
relaxation as well as cross relaxation and energy transfer
upconversion, which are a source of heat generation in the system.
In the Yb3+-doped samples, which have only two manifolds, energy
transfer upconversion is absent. That is the reason why the Yb3+ ions
are the most suitable dopants for laser cooling of solids.
Let us consider multiphonon relaxation. In this case the relaxation
process between upper and lower electronic states includes both an
electronic transition and a vibrational transition with the emission
of one or several phonons. The combined energy of these phonons
has to be equal to the energy gap between the levels involved in the
transition. Contrary to high-order multiphoton processes, which are
extremely weak, multiphonon processes can be relatively strong
since the electron–phonon interaction is stronger than electron–
electron interaction and phonons have a density of state that
typically are 10 orders of magnitude larger than that of photons.
The large variation in vibrational spectra of hosts makes the rate of
nonradiative relaxation strongly dependent on the host material. For
each host material all levels with the same energy gap have roughly
the same nonradiative transition rate, which relatively independent
on the nature of the electronic states involved in transition or the RE
ions involved in the process, unless a strong selection rule is involved.
The theory of multiphonon relaxation for RE ions was developed by
Kiel [29] and extended by Riseberg and Moos [30]. The nonradiative
decay rate, Wnr, is inversely proportional to the exponential function
of the energy gap separating the two levels (the energy gap law) [30]
Wnr = b e -a M (2.28)
where M is the number of emitted phonons. α and β are vibrational
and electronic parameters of the host material, respectively. As
one can see in Eq. 2.28 the nonradiative decay rate, Wnr, decreases
exponentially with an increasing number of phonons, M, participating
in the decay process. The vibrations with highest energy, Ωmax,
that is, optical phonons, provide the highest multiphonon relaxation
rate. The maximum phonon energy depends on the host material,
for example, Ωmax ≈ 203 cm–1 for a KPC crystal and Ωmax ≈ 1400
cm–1 for borate glasses. Since nonradiative decay is a source of heat
generation in the system only RE-doped hosts with Wnr << Wr are
suitable for laser cooling. The radiative relaxation rate of the first
excited state of RE ions is in the range Wr = 102–10
3 s–1 for a wide
3 –1
variety of hosts. Assuming Wr ≈ 10 s and using Eqs. 2.24 and 2.28
one can estimate that the emission quantum efficiency ηq > 0.99 can
only be reached in the samples with Ωmax ≈ < ΔE01/8 [31]. Here
ΔE01 is the energy gap between the manifolds involved into the
cooling cycle.
As one can see in Eq. 2.27, if the system is pumped at longer
wavelengths the cooling efficiency increases. However, as one moves
toward longer wavelengths the absorption cross section decreases
and the pump power absorbed by the sample decreases too. For this
reason, a figure of merit, Fcool(λ), for the cooling process has to be
considered.
Fcool(λ) = σabs(λ)ηcool(λ) (2.29)
The maximum Fcool(λ) can be used to evaluate the host material
for the laser cooling application.
2.3.2 Reabsorption or Fluorescence Trapping and Total

Internal Reflection
The ions in a sample can interact with each other. One of these mutual
interactions is reabsorption, when energy transfer between the ions
is radiative (Fig. 2.3a). In this case real photons are emitted by an
excited ion are absorbed by another unexcited ions within a photon
travel distance. This cycle can repeat itself several times resulting
in long-range energy diffusion between identical ions and gives
rise to the radiation trapping effect. The probability of this process
is proportional to R–2, where R is the distance between ions [32].
Let us assume that an emitted photon undergoes N reabsorption/
reemission events and reaches the boundary of the sample at the
angle smaller than the critical angle for TIR θc = sin–1(1/n), where n is
the refractive index of the sample. Following Snell’s law this photon
leaves the sample and do not undergo to TIR at the boundary. The
probability of reabsorption for this photon is r = 1 – exp(–αrL). Here
L is the average distance that photons travel to reach the boundary
Ú
of the sample. a r = a abs (n )F0 (n )dn is the average reabsorption
coefficient, which can be defined as the integrated product of the
absorption spectrum, αabs(ν), and the normalized fluorescence
intensity, F0(ν), here Ú F (n )dn = 1 . The fraction of emitted photons
0
that after reabsorptions leave sample is η1 = 1 + exp(αrL) [33]. Thus

in the trapping conditions provided reabsorption the fluorescence
lifetime increases and becomes equal to
tf = t f (1 + a r L) (2.30)
(a) (b) (c)

Figure 2.3 Energy transfer processes between two ions: (a) radiative
transfer, (b) nonradiative transfer, and (c) cooperative
emission.
As one can see in Eq. 2.30 the fluorescence lifetime, tf , increases
with the increase in the ion concentration and the size of the sample.
The fraction of photons, which leaves the sample and do not undergo
TIR is Pfree = 1 – cos(θc) = 1 – [1 – (1/n)2]1/2. For example, for the
Yb3+:ZBLAN sample with the refractive index n = 1.49 Pfree = 25.9%,
and for the Yb3+:YAG sample with the refractive index n = 1.833 Pfree
= 16.29%. Most of the anti-Stokes emitted photons undergo TIR. The
influence of TIR and reabsorption on solid-state optical cooling has
been investigated in details in Ref. [33] with a random walk model
developed in Refs. [34, 35]. The total escape efficiency or external
quantum efficiency in the system, where reabsorption and TIR are
taken into account, is
(hq )N {exp( -a r L)/ n - [1 - (1 n)2 ]1 2 + 1}
hext = (2.31)
1 - {[1 - (1 / n)2 ]1 2 - exp( -a r L) n}(hq )N
where N = G(L/lave)2 is the number of emission effects, lave is the
step size or mean free path, that is, the distance corresponding to
a probability for reabsorption of one half, that is, exp(–lave·αr) = 0.5.
Here G = 0.308·(lave/L) + 1.078·(lave/L)2 is the dimensionless scaling

factor estimated in Ref. [33].
The fluorescence lifetime described with Eq. 2.30 increases with
the increase in the ion concentration, but as the ion concentration
reaches some critical value, N0, the fluorescence lifetime begins
to decrease. This dependence has been experimentally observed
in a number of the Yb3+-doped samples [36]. The decrease in the
fluorescence lifetime of the excited Yb3+ ions with the increase in
the ion concentration indicates that the additional nonradiative
processes, which depend on ion concentration, begin to play
an important role as the ion concentration exceeds its critical
value, N0, which is different for different host materials. As the
ion concentration reaches its critical value two ions can interact
nonradiatively with energy transfer. The excitation jumps from
one ion to another before one is able to decay with a photon
generation and the first ion returns to the ground state (Fig. 2.3b).
The mutual interactions of two ions with nearly equal energy gap
between the ground and the first excited manifolds are Coulomb
interactions of the van der Waals type [37]. Using the quantum
mechanical theory Förster considered the dipole–dipole interaction
assuming that the interaction between the ions is strongest if for
both transitions electric–dipole transitions are allowed. In this case
the energy transfer probability is proportional to the inverse of
the sixth power of the interaction distance that is R–6. Dexter has
developed the theory of ion–ion energy transfer, including higher
multipole and exchange interactions [38]. Following Dexter’s theory
the energy transfer is much more likely in a system in which there
is significant overlap between the excited state transitions of the
donor ions and the ground-state absorptions of the acceptor ions.
At the critical distance between the ions, R0, energy transfer and
spontaneous decay processes have equal probability [32]. Assuming
that the energy exchange between ions takes place via dipole–dipole
interaction, the total concentration dependent fluorescence lifetime
is given by [39, 40]
t (1 + a r L)
tf ( NT ) = f 2
9 Ê NT ˆ (2.32)
1+
2p ÁË N0 ˜¯
where τf is the fluorescence lifetime in absence of excitation

-1
Ê4 ˆ
migration. N0 = Á p R03 ˜ is a critical ion concentration. For example,
Ë3 ¯
for the Yb3+:YAG sample N0 = 2.3 × 1021 cm–3. Migrating excitation
energy can reach an impurity ion, where it relaxes nonradiatively
with phonon (heat) generation. The energy migration process forms
an additional loss channel in the system, which is undesirable for the
cooling process. It has to be avoided especially in the relatively “big”
millimeter- or centimeter-size samples [41].
2.3.3 Cooperative Emission

The cooperative de-excitation of two adjacent excited ions can also
take place at high ion concentration in the sample. In this case two
simultaneously excited Yb3+ ions located nearby to each other relax
with emission of a visible photon (Fig. 2.3c). The cooperative emission
lifetime is nearly half of that of the infrared (IR) luminescence,
τf [42]. The contribution into cooling power from cooperative
hc 1
emission can be described as Pcoop = XN12 , where X is the
lcoop NT
experimental parameter [42]. The intensity of this radiation is very
weak in comparison with spontaneous fluorescence [41].
2.3.4 Achievable Temperature

The final equilibrium temperature of the sample depends on the
environment. Since most laser cooling experiments have been
performed in vacuum chambers, the same conditions are applied
here. Hence, there is only a radiative heat load in the system and
the final equilibrium temperature of the sample can be estimated
following the well-known Stefan–Boltzmann law
( )
VSamp Pcool = es b SSamp Tr4 - Ts4 (2.33)
where Tr is room temperature, Ts is the equilibrium temperature

of the sample, ε is the hemispherical emissivity of the sample, and
σB = 5.67 × 10–12 W·cm–2·K is the Stefan–Boltzmann constant. VSamp
and SSamp are the volume and the total surface area of the sample,
respectively. Let us following Ref. [41] consider the geometrical

efficiency factor (GEF) of the sample as
ηSamp = VSamp/SSamp (2.34)
Of course, a sphere is the best choice for cooling due to its high
volume-to-surface ratio. Indeed, for a sphere sample ηSamp = RSamp/3,
where RSamp is the radius of the sphere. For the cube, ηSamp = LSamp/6.
Using Eqs. 2.33, 2.23b, 2.24, and 2.25 one can obtain the equation
È 1 n ˘
Tr4 - Ts4 = hSamp Íhext habs f - 1˙ (a r + a b )Ip (2.35)
ÍÎ es B np ˙˚
Following Eq. 2.35 we can conclude that if one wants to reach
some predetermined temperature, Ts, in samples with equal ion
concentrations but different sizes, the pump intensity, Ip, has to
be chosen carefully. It should be noted that not all predetermined
temperatures can be reached using a sample with an arbitrary size
and ion concentration. The threshold GEF of the sample for any
predetermined final equilibrium sample temperature, Ts, can be
easily estimated. Let us assume that the pump intensity is equal
to the saturation pump intensity. In this case the absorbed power
density in the sample is Pabs ª NTs a (n p )IpSat and the threshold GEF
of the sample can be estimated as
hSamp
min
ª
(
es B Tr4 - Ts4 ) (2.36)
hcools a (n p )NT IpSat
Figure 2.4 illustrates the dependence of the threshold GEF of the
sample, hSamp
min
, simulated with Eq. 2.36 on the ion concentration if
the final equilibrium temperature of the sample is Ts = 80 K or 100 K.
2.4 Optical Cavities
2.4.1 Intracavity Cooling

Laser cooling of solids is the only solid-state technology, which can
cool to cryogenic temperatures. Unfortunately, the efficiency of this
cooling process is very low. Indeed, in RE-based optical refrigeration
the cooling power is only several percent of the absorbed pump
Optical Cavities 55
power. As one can see in Eq. 2.33 the temperature of the sample, Ts,
decreases as the cooling power, Pcool, increases. The cooling power
Pcool = –ηcool∙Pabs is proportional to the absorbed pump power. The
pump power absorbed in the sample on the length Labs is
Pabs = Pp (1 - e -a abs Labs ) (2.37)
where αabs = αr(ν, T) + αb is the absorption coefficient. Due to

Boltzmann distribution of electrons in the ground and excited
manifolds, resonant absorption decreases with the decreasing
temperature. The decrease in the resonant absorption, αr(ν,T), can
be compensated by the increase in the path length Labs. In 1968,
Kushida and Geusic placed the Nd3+:YAG cooling sample inside the
cavity of the Nd3+:YAG laser [11]. In an attempt to cool a Nd3+:YAG
crystal for the first time they observed a reduction of heating. The
intracavity technique has the advantage of very high intracavity
circulating power (Fig. 2.5). In 2004, this technique was used by
Heeg and his colleagues to realize the neat laser cooling process
in a Yb3+:ZBLAN sample with the temperature drop 6 K from room
temperature.
hSamp
min
Figure 2.4 The dependence of ion concentration on the threshold GEF,

hSamp
min
, of Yb3+:YAG if Ts = 80 K and 100 K.
Figure 2.5 An intracavity cooling scheme: a cooling sample is placed

inside a laser cavity.
In 2014, this method was applied to a vertical external cavity

surface-emitting laser (VECSEL) [43]. In about 6 min a 7% Yb3+:YLF
sample placed in the cavity of VECSEL was cooled from room
temperature down to 131 K.
2.4.2 Nonresonant Cavity

In 1998, Edwards et al. proposed to use the nonresonant cavity
technique to enhance the absorption in the system [44, 45]. In the
nonresonant cavity technique the pump beam travels through the
cooling sample placed in the resonator at an angle slightly off normal
to the mirrors (Fig. 2.6). In Edwards’s experiment a Yb3+:ZBLAN
cylinder with 12 mm diameter and 10 mm long was mirrored on
both ends (Fig. 2.6a). One mirror had a pinhole with diameter 150
μm for admitting the pump radiation.
(a) (b)
Figure 2.6 Nonresonant cavities formed by dielectric mirrors deposited
on the sample (a) and by external mirrors (b).
A maximum temperature drop 48 K starting from room

temperature was achieved. This technique has a number of
drawbacks. Indeed, besides pump tapping the mirrors trap the
fluorescence increasing the reabsorption in the system. Mirror
heating is undesirable for the cooling process. In 2003, Hoyt et al.
proposed to use external mirrors (Fig. 2.6b) [46]. The improved
Laser Cooling in Rare Earth–Doped Glasses and Crystals 57
technique permits to reduce fluorescence trapping and eliminate

heat conduction between mirrors and a sample. In this experiment a
1 wt% Tm3+:ZIBLANP sample of volume 4 ¥ 4 ¥ 8 mm3 was placed in
the nonresonant cavity with external dielectric mirrors, which had
reflectivity ~99.9%. A planar mirror had a ~0.4 mm hole to admit
the pump beam. A second mirror had a 1 m radius of curvature. A
temperature change of 19 K from room temperature is achieved in
this technique. The nonresonant external cavity technique was used
in the experiment where a record temperature drop (ΔT = 185 K)
was achieved in a Yb3+:YLF sample [17].
2.4.3 Resonant Cavity

In 2003, a resonant cavity was proposed to enhance pump light
absorption [46]. It has been shown that theoretically pump light
absorption can reach unity for the mirrors with the reflectivity
1 and exp(–2αabsLR), where αabs = αr(ν,T) + αb. Since the resonant
absorption αr(ν,T) is a function of temperature the length LR has to
be properly arranged with temperature. A piezoelectric stabilization
can be a solution to the problem. As one can see in Refs. [47–49]
the absorption experimentally achieved with the resonant cavity
is comparable with the absorption achieved with the nonresonant
cavity technique. It reaches ~90%.
2.5 Laser Cooling in Rare Earth–Doped Glasses

and Crystals
The properties of radiative transitions can be investigated with
absorption and emission spectra. All spectral lines can be divided
in three major types of line shapes: Lorentzian, Gaussian, and
Voigt. Any process, which has the same probability for all ions of
the system, is a source of a Lorentzian line shape, gL. All physical
processes with a random distribution of occurrence for each ion
are a source of a Gaussian line shape, gG. The Voigt line shape, gV,
is a result of the convolution of two broadening mechanisms: one
of which results in a Gaussian broadening and the other produces a
Lorentzian profile. Lorentzian broadening is referred to as lifetime
broadening since the processes responsible for this type of spectral
line shapes are generally the processes that shorten the lifetimes
of the energy levels involved in the transition, as an example, the
Heisenberg uncertainty relationship relating time and energy.
This contribution can be considered as the natural linewidth for a
transition. The homogeneous spectral line shape can be described
with the area-normalized Lorentzian profile
2Dn L
gL (n ,T ) = (2.38)
p ÈÎ( Dn L ) + 4(n - n0 )2 ˘˚
2
where ν0 is the mean transition frequency (Fig. 2.7a). ΔνL is the full-
width at half-maximum (FWHM) given by the relation [50, 51]
1 1
Dn L = + (2.39)
2pt r pT2
where T2 is the dephasing time. As the temperature decreases the
homogeneous transition linewidth ΔνL is observed to narrow [53].
At very low temperatures (up to 40 K), the homogeneous linewidth
is observed to follow a linear increase law with temperature (i.e.,
ΔνL ≈ T) at higher temperatures (up to 300 K), the increase is nearly
quadratic (i.e., ΔνL ≈ T2) [51, 52]. The one-phonon direct broadening
is associated with liner temperature dependence; the two-phonon
Raman scattering is associated with a T2 low at higher temperatures
[54]. Each dopant ion occupies a unique site in the host material and
experiences a unique Stark splitting. The structural disorder in a
host is a source of inhomogeneous broadening, which is generally
well described by a Gaussian distribution (Fig. 2.7b). In a perfect
(unrealistic) crystal all dopant ions are subjected to an identical
crystal field and as a result are free from inhomogeneous broadening.
(a) (b)
Figure 2.7 Homogeneous broadening (a) and inhomogeneous broadening
(b).
On the other hand glass hosts with their structural disorder

provide different coordination environment for dopant ions, which
results in inhomogeneous broadening. Figure 2.7a illustrates
homogeneous absorption at several ion concentrations and Fig. 2.7b
illustrates inhomogeneous absorption at one concentration by a
number of subpopulations. As one can see in Fig. 2.7 if one pumps a
sample with the homogeneous broadening at the long wavelength tail
of the absorption spectrum all ions participate in cooling. If a sample
with inhomogeneous broadening is pumped at the long wavelength
tail of the absorption spectrum some ions can participate in the
heat generation process. Inhomogeneous linewidths are essentially
independent on temperature. The temperature dependence of the
homogeneous and inhomogeneous broadening in the Yb3+:ZBLANP
glass sample has been investigated in Ref. [55], where the Voigt
absorption and emission profiles have been deconvoluted. The
inhomogeneous linewidth for the Yb3+:ZBLANP sample is ΔvG ≈ 990 ±
40 GHz. It is almost independent on temperature. The homogeneous
linewidth for the Yb3+:ZBLANP sample changes with temperature as
ΔvL ≈ (12 ± 2) × 10–3T(1.9±0.1) in the temperature range of 10–300 K.
It is well known that the resonant absorption coefficient, αr(ν,T),
can be estimated as a product of the absorption cross section σa(ν,T)
and the thermal population density, N0(E0i,T), of the initial state
with the energy E0i that is αr(νp,T) = σa(νp,T)∙N0(E0i,T). Following
the Boltzmann distribution described with Eqs. 2.14 and 2.15 the
thermal population density of the initial state N0(E0i,T) ≈ exp(–
E0i/kBT). The absorption cross section is proportional to the Voigt
spectral line that is σa(νp,T) ≈ gV(ν,T). Consequently, the resonant
absorption coefficient can be presented as
αr ≈ gV(ν,T)∙exp(–E0i/kBT) (2.40)
As one can see in Ref. [55] the inhomogeneous broadening is
dominant for glass hosts. It almost do not change with temperature
that is gV(ν,T) ≈ constant. As one can see in Eq. 2.40 the peak
absorption coefficient decreases exponentially with temperature in
glass hosts. In crystal hosts homogeneous broadening is dominant.
The homogeneous linewidth scales as ΔνL ≈ T2 in the temperature
range 40–300 K, which is interesting for optical refrigeration. As
one can see in Eq. 2.38 the peak value gL(ν0,T) ≈ T–2 at ν = ν0. In
this case from Eq. 2.40 the resonant absorption coefficient αr ≈
T–2∙exp(–E0i/kBT). That is in crystal hosts temperature partially

offsets the decrease in thermal population of the level. Thus one
can conclude that with decreasing temperature a peak absorption
coefficient αr(ν0,T) in crystal hosts exceeds one in glass hosts. The
absorption efficiency described with Eq. 2.25 can be considered as
a “competition” between the resonant absorption coefficient αr and
the background absorption coefficient αb. Indeed, ηabs = αr/(αr + αb),
that is, the higher peak absorption in crystal hosts, αr, permits to
tolerate the higher background absorption coefficient αb in the host
material. That makes crystal hosts more suitable for laser cooling of
solids.
Tremendous progress in purification of fluoride glasses in 1990s
caused by development of long-haul fiber-optic telecommunication
links permitted to achieve the net laser cooling process in a
Yb3+:ZBLANP glass sample for the first time in 1995 [13]. The
advantages of RE-doped crystals over RE-doped glasses for laser
cooling to low temperatures for the first time was demonstrated by
Bowman and Mungan in 2000 [57]. In this work local laser cooling in
the Yb3+-doped crystal samples (Yb3+:KGW, Yb3+:YLF, and Yb3+:YAG)
has been demonstrated for the first time. Net laser cooling in a
Yb3+:YAG crystal sample for the first time was demonstrated in 2001
by Epstein’s research team [58]. All latest record temperatures have
been achieved in the Yb3+:YLF crystal samples [14, 16]. Let us briefly
consider the most well-known experiments in which net laser
cooling of solids has been demonstrated.
2.5.1 Laser Cooling in Rare Earth–Doped Glasses
2.5.1.1 Rare earth–doped fluoride glasses
2.5.1.1.1 Yb3+:ZBLAN and Yb3+:ZBLANP fiber and bulk

samples
Heavy metal fluoride glasses including a family of glasses with a
composition ZrF4-BaF2-LaF3-AlF3-NaF (ZBLAN) were developed
by Poulain and Lucas at the University of Rennes in France in 1975
[59]. ZBLAN has several undisputable advantages over other glass-
es: low intrinsic losses (theoretically 0.001–0.01 dB/km), the wide
transparency window (0.35–4 μm), and the small maximum pho-
non energy (~580 cm–1) [60–62]. Extrinsic absorption originates

from impurities such as transition metals (TMs) (Fe, Cu, Co, Ni),
rare earths (Nd, Tb, Er), OH and H2O, and other molecular species
(CO, CO2, NH4+). Extrinsic scattering occurs as a result of bubbles,
crystals inclusions, geometrical defects, and other imperfections.
The refractive index of ZBLAN is n ≈ 1.5. It has negative thermo-
optic coefficient dn/dT = –14.75 × 10–6 (K–1). Its thermal conductiv-
ity is ~0.9 W/(mK) and expansion coefficient is 17.2 × 10–6 (K–1).
ZBLAN has a relatively low glass transition temperature Tg ≈ 260°C.
Fluoride glasses have a number of drawbacks: fragility and sensi-
tivity to acids. In a number of experiments devoted to laser cooling
of solids a glass composition 53%ZrF4-18%BaF2-3%LaF3-3%AlF3-
20%NaF-2%PbF2-1%YbF3 (Yb3+:ZBLANP) have been used. Heavy
Pb permits to reduce the maximum phonon energy down to ~506
cm–1 and consequently to increase the emission quantum efficiency
as one can see in Eq. 2.24. A number of successful laser cooling ex-
periments, in which net laser cooling has been achieved, have been
done in bulk as well as in fiber RE-doped ZBLAN samples. In the
very first proof of concept experiment a bulk 2.5 × 2.5 × 6.9 mm3
ZBLANP sample doped with 1 wt% of Yb ions was cooled down by
0.3 K below room temperature [13]. The pump beam at λp = 1030
nm from a continuous-wave titanium–sapphire laser was directed
along the long axis of the sample in a single pass geometry. The mean
fluorescence wavelength was λf = 995 nm. The experimental cooling
efficiency was ~2%. The temperature of the sample was monitored
with a liquid-nitrogen InSb IR camera [13]. The largest net laser
cooling in the bulk geometry in a cylindrical fiber preform with di-
ameter 8 mm and length 8 mm was achieved in 2005 by Thiede [63].
A 2% Yb3+-doped ZBLANP sample was cooled by ~92 K from room
temperature. It was pumped at wavelength 1026 nm with a 10 W
Yb3+:YAG laser. The most notable advance with Yb3+:ZBLAN fiber
samples was achieved by Gosnell in 1999, when a 1 wt% Yb3+-doped
ZBALN multimode fiber with diameter 175 μm and length 7 mm was
cooled by 65 K from room temperature in a double-pass geometry
pumped at wavelength 1015 nm with a 2.2 W pump source [28].
2.5.1.1.2 Tm3+:ZBLANP bulk samples

In 2000 and 2003, laser cooling with Tm3+ ions doped in ZBLANP
was demonstrated. In Tm3+:ZBLANP the ground, 3H6, and the first
excited, 3F4, manifolds of Tm3+ ions were involved in the cooling

cycle [46, 56]. The energy gap between the 3H6 and 3F4 manifolds
is ~6000 cm–1. In Ref. [56] several Tm3+-doped glass and crystal
samples such as CaF2, BaF2, YAlO, LuAG, and ZBLANP have been
investigated on the subject of cooling with pump excitation at 1.82
μm < λp < 1.97 μm. But only with a Tm3+:ZBLANP sample the net
laser cooling effect was achieved. The sample with volume 5 × 5 × 9
mm3 consisted of high purity 1 wt% Tm3+:ZBLANP was cut from a
fiber preform. It was pumped with light from an optical parametric
oscillator (OPO) at wavelength 1900 nm with the incident
average pump power of ~3 W. A calibrated Mach–Zehnder (MZ)
interferometer was used to monitor the temperature of the sample.
The mean fluorescence wavelength was λf = 1820 nm. The energy
gap between the manifolds of Tm3+ ions involved into the cooling
cycle is less that the energy gap between the manifolds in Yb3+ ions.
That means the nonradiative decay rate, Wnr, in Tm3+ ions exceeds
the nonradiative decay rate in Yb3+ ions (see Eq. 2.28). As seen from
Eq. 2.24 the increase in the nonradiative decay rate is a source of the
decrease in the emission quantum efficiency, ηq. In Tm3+:ZBLANP
the nonradiative decay rate Wnr ≈ 0.64 s–1, the radiative decay rate
Wr ≈ 83 s–1, the emission quantum efficiency ηq = 99.2%. The sample
was cooled by 1.2 K from room temperature for a single pass of the
pump beam. In spite of the increase in the nonradiative decay rate in
Tm3+:ZBLANP samples the cooling efficiency of Tm3+:ZBLANP, which
is ηcool ≈ 4.8%, exceeds the cooling efficiency of Yb3+:ZBLANP, which
is ηcool ≈ 3.5%. But Tm ions have several manifolds. They are not free
from excited-state absorption (ESA), which deteriorates the cooling
process. In 2003, a 2 wt% Tm3+:ZBLANP sample of 3 × 3 × 10 mm3
volume was cooled by 19 K from room temperature in a single-pass
geometry. The sample was pumped at λp = 1900 nm with a pump of
~2.72 W [46].
2.5.1.1.3 Tm3+:BIG bulk samples

The resonant absorption coefficient αr ≈ σa(νp)NT is proportional
to the ion concentration NT. As one can see in Eq. 2.25 the increase
in αr permits one to reach higher absorption efficiency ηabs in the
samples with the identical background absorption coefficient αb.
However, crystallization is an obstacle to fabricate ZBLAN samples
with high ion concentration. 10 mol% is the maximum concentration

of RE ions in ZBLAN without inducing detrimental crystallization.
The composition BaF2-InF3-GaF3-ZnF2-LuF3-GdF3 (BIG) can be
considered as a solution to the problem, since it can have higher
concentration of Yb3+ ions than ZBLAN. In 1999 a Yb3+:BIG sample
was investigated by Murtagh et al. [64]. Theoretical cooling
efficiencies indicate that BIG hosts outperform ZBLANP hosts below
80 K by a factor of 2. BIG may attain a minimum temperature of 45
K compared to 55 K for ZBLANP [64]. The laser cooling process in
Yb3+-doped BIG samples has been investigated by Fernandez et al.
[65, 66].
2.5.1.2 Laser cooling in rare earth–doped fluorochloride

glasses: Yb3+:CNBZn and Er3+:CNBZn bulk samples
The chloride ions in fluorochloride composition such as CdF2-
CdCl2-NaF-BbF2-BaCl2-ZnF2 (CNBZn) permit to reduce the
maximum phonon energy down to ~370 cm–1 and thus to reduce
the multiphonon relaxation rate. The reduced maximum phonon
energy of a host is very important for laser cooling with RE ions if
the energy gap between the manifolds involved in the cooling cycle
reduces, like in the case of Tm3+ or Er3+ ions. In a number of works
nonhygroscopic CNBZn glass samples doped with Yb3+ ions have
been investigated for laser cooling applications [65, 66, 67, 68]. A
CNBZn sample of volume 2 × 8 × 10 mm3 doped with 1 mol% of Yb3+
pumped at λp = 1010 nm demonstrated the cooling efficiency ~25%
at 300 K, which falls about 20% at 77 K [65]. In 2006 Fernandez et
al. observed a temperature drop 0.5 K in a bulk-cooling experiment
in a 10.7 × 10.7 × 2.2 mm3 sized Er3+: CNBZn sample pumped at 860
nm with a pump of 1.9 W [69]. The mean fluorescence wavelength
of a Er3+:CNBZn sample was 852.5 nm. The cooling efficiency was
0.7%. The cooling process was monitored by using a combination of
photothermal deflection (PTD) measurements and time evolution of
the average temperature of the sample acquired with an IR camera.
Table 2.2 summarizes the results of experiments devoted to laser
cooling of RE-doped glasses in which net laser cooling of solids
has been achieved. Figure 2.8 illustrates progress, which has been
made toward laser cooling in Yb3+-doped glasses since the very first
successful experiment in 1995.
Table 2.2 Net laser cooling with RE-doped glasses
Host MPE Nc ΔT λp Sample Reference

Yb3+
FIBER
ZBLANP 506 1 wt% 16 1015 Dco = 250, L = 10 [74, 75]
ZBLANP 1 wt% 21 1015 Dco = 175, L = 10 [27]
ZBLANP 1 wt% 65 1015 Dco = 175, L = 7 [28]
ZBLAN 580 1 wt% 13 1015 Dco = 175, L = 40 [76]
ZBLAN 1 wt% 3.7 1013 Dco = 175 [93]
BULK
ZBLANP 1 wt% 0.3 1015 V = 2.5 × 2.5 × 6.9 [13]
ZBLAN 48 1030 D = 12, L = 10 [45]
ZBLAN 2% 6.0 1027 V = 2 × 2 × 3 [77]
ZBLANP 2 wt% 92 1026 D = 8, L = 8 [63]
Tm3+
ZBLANP 1 wt% 1.2 1900 V = 5 × 5 × 9 [56]
ZBLANP 1 wt% 19 1900 V = 4 × 4 × 8 [46]
ZBLANP 2 wt% 2 1900 V = 3 × 3 × 10 [46]
Er3+
CBNZn 370 0.5% 0.5 860 V = 10.7 × 10.7 × 2.2 [69]
MPE (cm–1), the maximum phonon energy; Nc: the ion concentration; ΔT (K), the
temperature drop; λp (nm), the pump wavelength; Dco (μm), the fiber core diameter;
D (mm), the preform diameter; L (mm), the length of the sample; V (mm3), the volume
of the sample.
2.5.2 Laser Cooling in Rare Earth–Doped Crystals
2.5.2.1 Yb3+-doped oxide crystal (Yb3+:YAG)

As already noted, the higher peak absorption cross sections in
crystal hosts than in glass hosts can be useful for laser cooling to low
temperatures. After a number of successful laser cooling experiments
with Yb3+-doped ZBLAN and ZBLANP glass samples, in 2001, net
laser cooling was achieved for the first time in a Yb3+:YAG crystal
sample [58]. A 2.3% Yb3+:YAG sample was pumped at wavelength
1030 nm with a pump of 1.8 W. YAG is the oxide crystal. It is non-

hygroscopic. YAG has greater thermal conductivity (~13 W∙m–1K–1)
than ZBLAN (~0.9 W∙m–1K–1). The refractive index (1.83) and the
maximum phonon energy (630 cm–1) of YAG exceed the refractive
index (1.5) and the maximum phonon energy (580 cm–1) of ZBLAN. As
one can see in Eq. 2.30 the increase in the refractive index is a source
of the increase in the fluorescence trapping, which is undesirable for
laser cooling, since it results in a decrease in the external quantum
efficiency. The increase in the maximum phonon energy results in a
decrease in the emission quantum efficiency, which is undesirable
for laser cooling too. The coefficient of thermal expansion of YAG
is 6.14 × 10–6 K–1, and its thermo-optic coefficient is dn/dt ≈ 8.4 ×
10–6 K–1. In 2001, the first experiment devoted to laser cooling of
Yb3+:YAG was done in vacuum therefore only the radiative heat load
took place. In 2013, a Yb3+:YAG sample was laser-cooled in air [15].
In this experiment Brewster-cut 3 at.% Yb3+:YAG crystal of 1 × 1 ×
10 mm3 volume was cooled by 8.8 K from room temperature with a
pump of 4.2 W at wavelength 1029 nm. The laser cooling process in
Yb3+:YAG has been theoretically considered in Refs. [41, 80].
Figure 2.8 The temperature drop, ΔT, in Yb3+-doped glasses achieved in

different years.
2.5.2.2 Er3+-doped chloride crystal (Er3+:KPC)

In 2006, net laser cooling in a Er3+-doped chloride KPb2Cl5 (KPC)
crystal sample with low maximum phonon energy (~203 cm–1) for
the first time was observed by Fernandez et al. [69]. KPC is only
slightly hygroscopic. A 4.5 × 6.5 × 2.7 mm3 sized 0.5 mol% Er3+-
doped KPC sample pumped with a pump of 1.9 W at 870 nm was

cooled by ~0.7 K starting from room temperature. As one can see
in Eqs. 2.28 and 2.24 low phonon energy can provide high emission
quantum efficiency even for RE ions with a small energy gap between
the manifolds involved in the cooling cycle like Ho3+ (5I8 – 5I7) and
Dy3+ (6H15/2 – 6H13/2). The laser cooling process in Ho3+:KPC and
Dy3+:KPC samples has been investigated in Refs. [70, 71].
2.5.2.3 Rare earth–doped fluoride crystals
2.5.2.3.1 Yb3+- and Tm3+-doped BYF crystals

In 2004, Patterson et al. observed laser cooling in a Tm3+-doped
fluoride BaY2F6 (BYF) crystal with low (~350 cm–1) maximum
phonon energy with the temperature drop ~1.5 K. A 2 wt%
Tm3+:BYF sample was pumped with a OPO (~6 W) at wavelength
1902 nm [72]. Four years later, in 2008, Patterson et al. cooled a 1.2%
Tm3+:BYF sample by 3.2 K from room temperature with a pump of
4.4 W at wavelength 1855 nm [73]. In 2006, Bigotta et al. observed
laser cooling in a 2.5 mol% Yb3+-doped 3 × 4 × 10 mm3 BYF sample
in single-pass geometry with the temperature drop 4 K starting from
room temperature [81, 82].
2.5.2.3.2 Yb3+-doped YLF crystal

In 2007, Bigotta et al. for the first time demonstrated laser cooling in
a 4 × 4 × 12 mm3 Yb3+-doped YLiF4 (YLF) sample [83]. They pumped
a 5% Yb3+:YLF sample with a pump of 3 W at wavelength 1025 nm
and observed cooling by 6.3 K starting from room temperature.
YLF is one of widely used laser crystals. Due to the similar size,
yttrium ions can be replaced with laser-active RE ions without
strongly affecting the lattice structure. The thermal conductivity
of undoped YLF crystal is 5.3 W∙m–1K–1 at room temperature. It
increases up to 24 W∙m–1K–1 at 100 K. YLF is a low-phonon crystal
with the maximum phonon energy ~350 cm–1. YLF is birefringent
with tetragonal crystal symmetry. It has 0.1–7.5 μm transition band.
YLF has the refractive index no = 1.448, ne = 1.47, and the negative
thermo-optic coefficient dno/dT = –4.6 × 10–6 K–1, dne/dT = –6.6 ×
10–6 K–1 at room temperature, which decreases down to dno/dT =
–0.5 × 10–6 K–1, dne/dT = –1.8 × 10–6 K–1 at 100 K. Since 2007, several
successful laser cooling experiments have been done with Yb3+:YLF
samples with Yb3+ concentrations in the range 1%–10% (Table 2.3).

In 2013, the first laser cooling to a cryogenic temperature of 119 K,
which is below the National Institute of Standards and Technology
(NIST) defined cryogenic temperature of 123 K, was achieved in a
5 wt% Yb3+:YLF crystal sample pumped at 1020 nm [59]. In 2015,
the first laser cooling to double digit temperature 91 K was achieved
in a 10% Yb3+:YLF sample pumped at 1020 nm [17]. The source of
this success is the improved purity of the host material (αb = 10–4
cm–1) and enhanced absorption of the pump light (42% over the
previous experiment). A record temperature 91 K was reached in
~10 min starting from ~255 K. Table 2.3 summarizes the results of
experiments devoted to laser cooling of RE-doped crystals in which
net laser cooling of solids has been achieved. The progress toward
laser cooling in Yb3+-doped crystals is illustrated in Fig. 2.9.
Figure 2.9 The temperature drop, ΔT, in Yb3+-doped crystals achieved in

different years.
Figure 2.10 illustrates the progress toward net laser cooling in

RE-doped glasses and crystals. As one can see in Fig. 2.10 net laser
cooling has been realized with the Yb3+, Tm3+, and Er3+ ions doped in
a number of low-phonon glass and crystal hosts. Figure 2.10 presents
a record temperature drop for each host material. The numbers in
the brackets illustrate the year when the result was achieved. All RE-
doped crystals presented in Fig. 2.10 are bulk ones with millimeter
sizes.
Table 2.3 Net laser cooling with RE-doped crystals
Host MPE Nc ΔT λp Sample References

Yb3+
YAG 630 2.3 wt% 8.9 1030 [58]
YAG 3% 8.8 1029 V = 1 × 1 × 10 [15]
BYF 350 2.5% 4 1025 V = 3 × 4 × 10 [81–83]
YLF 350 5% 6.3 1025 V = 4 × 4 × 12 [83]
YLF 5% 69 1030 [48]
YLF 5% 143 1022 [49]
YLF 5% 145 1023 [95]
YLF 5% 162 1020 V = 3 × 4 × 12 [14]
YLF 10% 185 1020 [16]
YLF 10% 174* 1020 [17]
Tm3+
BYF 2 wt% 1.5 1902 V = 4.5 × 4.5 × 7 [72]
BYF 1.2% 3.2 1855 [73]
Er3+
KPC 203 0.5% 0.7 870 V = 4.5 × 6.5 × 2.7 [68, 69]
KPC 3.6 × 0.0136 1567.6 D = 3, L = 13.6 [94]
10–19 cm–3
MPE (cm–1), the maximum phonon energy; Nc, the ion concentration; ΔT (K), the
temperature drop; λp (nm), the pump wavelength; V (mm3), the volume of the sample;
D (mm), the diameter of the sample; L (mm), the length of the sample; *, the double-
digit temperature (91 K) has been reached.
Figure 2.10 The progress toward net laser cooling in RE-doped glasses and
crystals with a record temperature drop for each host material.
Temperature Measurements 69
2.5.3 Laser Cooling in Rare Earth–Doped Oxyfluoride

Glass Ceramic
RE-doped glass ceramic (GC) has a unique combination of high
chemical and mechanical stability related to the oxide glass and low
phonon energy of the fluoride nanocrystals, which trap the majority
of RE ions. The effective embedding of RE ions in the crystalline
phase with low phonon energy provides high quantum efficiency for
the RE ion transitions involved in cooling cycles. On the other hand,
GCs are much easier to manufacture on a large scale than are single
crystals. Laser cooling with Tm3+-doped oxyfluoride GCs has been
considered in Ref. [84]. Yb3+-doped GCs for optical refrigeration have
been investigated in Ref. [85]. In this work oxyfluoride glass with
chemical composition 30SiO2-15Al2O3-27CdF2-22PbF2-4YF3-2YbF3
was prepared by conventional melt-quenching technique. High-
purity (Sigma-Aldrich, 99.99%) precursors were well mixed in an
agate mortar to obtain homogeneous powder, and were transferred
to a covered platinum (Pt-100%) crucible. The crucible was fired at
1100°C for 2 h in a muffle furnace. The resulting molten glass was
poured into a stainless steel mold that was preheated at the glass
transition temperature, where the glass was annealed at the same
temperature for 5 h to remove internal stresses and strains. The glass
samples were annealed at different temperatures and timescales to
transform them to GCs. The Yb3+ glass and GC samples have been
pumped at 1020 nm [85]. Net laser cooling has not been achieved,
but a temperature rise of 40 K/W has been observed in the glass
samples when pumped at 1020 nm, a level that is not observed in
the Yb3+:GC samples even at the pump wavelength 1000 nm, where
the resonant absorption is much higher. That indicates a better ratio
of anti-Stokes emission to nonradiative decay in the GC samples
compared to the glass samples, which is important for cooling.
2.6 Temperature Measurements

The temperature of a sample can be measured with different
techniques, which can be divided into contact and noncontact ones.
These techniques can provide a local temperature measurement as
well as the bulk temperature of a sample. All these techniques have
some advantages and disadvantages and have to be properly chosen

for each experiment.
2.6.1 Contact Technique
2.6.1.1 Thermocouples
In 1968, Kushida and Geusic used a thermocouple to observe heat
mitigation in a Nd3+:YAG sample. It was attached to the sample. Later
a direct measurement of the temperature of the entire sample well
insulated in an evacuated chamber has been done with a fine wire
thermocouple in a number of experiments [45, 63, 76]. Conduction
of heat through the thermocouple wires and absorption of the
fluorescence are obstacles on a way high-quality measurement,
which can be avoided with noncontact techniques.
2.6.1.2 Fiber Bragg grating

In 2013, fiber Bragg grating was used in the first experiment devoted
to laser cooling of Yb3+:YAG in air to measure its temperature [15].
The Bragg wavelength shift was 9.9 pm/K. With a Bragg reflection
bandwidth of 616 pm, the total measurement range was 62 K. The
fiber Bragg grating was in contact with the sample. The fiber Bragg
grating was transparent to the fluorescence and the pump radiation.
This technique is free from cumbersome calibration and optical
alignment. Its dynamic range can be easily switched by changing the
interrogation system while keeping the grating on the sample.
2.6.2 Noncontact Technique
2.6.2.1 Photothermal deflection

Photothermal deflection (PTD) is noncontact thermometry
technique for analyzing small temperature changes in a sample. PTD
was developed in 1980 as a means of measuring optical absorbance
down to αabsl = 10–8 [86, 87]. In laser cooling of solids PTD is
employed to measure the local temperature gradients induced by the
pump lasers inside samples. In this case a pump beam is focused into
the sample and a probe beam passes through the sample in opposite
direction. Angular deflection of the probe beam, which is caused by
Temperature Measurements 71
thermally induced refractive index gradient in the pumped volume

of the sample, is measured with a position sensitive photodetector.
PTD has been used in a number of works to measure the temperature
of laser-cooled samples [57, 64, 65, 69, 74, 75]. Unfortunately, PTD
have a number of drawbacks. It can only be utilized to single-pass
geometries, since it requires careful alignment of the pump and
probe laser beams. It is inherently limited to probing one small
region of a sample defined by the probe laser. In this method it is
important to separate the thermal contribution to the refractive
index change (thermal lensing) from the accompanying contribution
caused by pump-induced population changes in both ground and
excited states (population lensing).
2.6.2.2 Thermal cameras

In 1995, a liquid-nitrogen-cooled InSb IR camera was used to
monitor optical refrigeration in a Yb3+:ZBLANP sample [13]. A
thermal camera (TC) has also been used in a number of experiments
devoted to laser cooling of solids to measure the temperature of
the sample [46, 56, 69]. It is worth to emphasize that TC can be
inappropriate for some samples (e.g., KPb2Cl5), because the sample
can be completely transparent in most of the mid-IR and is therefore
invisible to the TC. The detection range of TCs is limited to no lower
than 250 K with an absolute resolution of ~0.2 K
2.6.2.2 Differential luminescence thermometry

Differential luminescence thermometry (DLT) is the most widely
accepted noncontact technique in recent years. It was used for the
first time in 1997 by Mungan et al. to measure the temperature in a
1 wt% Yb3+:ZBLANP fiber sample [74]. The luminescence spectral
profile of a RE-doped sample depends on the temperature and
does not depend on the pump wavelength. DLT uses temperature-
dependent spectral change of a sample to achieve high temperature
sensitivity. The temperature of the sample can be determined by
direct comparison of sample spectra with a set of reference spectra
captured with a closed-cycle cryostat. The sample temperature
is then deduced from the reference spectra by a least-squares
comparison. DLT has been used in some works [28, 27, 74, 75, 76,
77], but due to acquisition of the full spectrum, this technique is
slow and in addition is not immune to inherent noise from CCD dark
current and electronic readout. The resolution of ~20–100 mK can
be reached with the DLT method.
2.6.2.3 Two-band differential spectral metrology

Two-band differential spectral metrology (2B-DSM) is based
on dual-band spectral differencing via a balanced amplified
photodetector (BAPD) directly in the spectral domain [78, 79]. Due
to the large electronic bandwidth and high common-mode rejection
of these detectors, a fast and shot-noise-limited performance can
be realized with this method. A monochromator or color filters can
be used to sample fluorescence spectra. This technique is sensitive
to any changes in the slope of the fluorescence spectrum caused by
temperature variations. 2B-DSM permits one to observe temperature
on the millisecond time scale.
2.6.2.4 Differential infrared thermography

Differential infrared thermography (DIT) overcomes some of
the limitations of the standard IR thermography setup. In IR
thermography background absorption is not eliminated and the
optical cooling channel competes with background absorption, which
is present even in undoped materials resulting in heat generation.
In the DIT technique two samples of the same host material are
located side by side. Both samples are irradiated simultaneously.
The laser spot has to be equally distributed among them. In this
way, the undoped sample can be used as a reference to subtract any
contributions to heating such as background absorption processes,
power fluctuations, etc. not related to laser cooling with the optically
active ions. DIT has been used in several experiments devoted to
laser cooling of solids [88, 89].
2.6.2.5 Polarization-resolved thermometry

Noncontact and sensitive techniques based on a change of reflectivity
[90] and thermally induced stress birefringence [91] in a balanced
detector arrangement can be used to determine the quantum and
cooling efficiencies. Polarization-resolved thermometry (PRT) is
based on thermally induced stress birefringence in a sample caused
by temperature gradients. It has a very high temperature resolution
Conclusions 73
of 5 mK. This technique holds certain advantages over existing

techniques at low starting temperatures.
2.6.2.6 Mach–Zehnder interferometry

An MZ interferometer has been proposed to measure the temperature
of a cooling sample [92]. The refractive index of the sample changes
with temperature over a wide temperature range. This approach has
been used by Hoyt et al. to measure the laser-induced temperature
change in a Tm3+:ZBLAN sample [46]. This technique can be applied
to single-pass and multipass geometries providing the bulk sample
temperature.
2.7 Conclusions
This chapter is focused on relatively comprehensive consideration
of the laser cooling process in RE-doped glasses and crystals. Laser
cooling of solids is interesting not only from the fundamental point of
view as a process of light and material interaction but its application
in portable space-based sensor systems as well. An understanding
of laser cooling of solids is very important for development of an
all-solid-state cryocooler, which is a compact, reliable device free
from vibrations. It is free from moving parts and fluids. It can be
used for cryogenic refrigeration of semiconductor devices in a
number of applications that have tight constraints for weight and
size, for example, airborne and space-based detector systems.
The operation of semiconductor detectors is based on electron
excitation. The incident photon excites electron from the valence
to conduction band inducing an electric current, which has to be
measured. But random thermal excitations (thermal noise) can also
excite electrons from the valence to conduction band resulting in the
“dark” current that limits the performance of detectors. Cryogenic
refrigeration of detectors can sufficiently reduce the thermal noise
and increase the detector sensitivity. Liquid helium, methane, and
ethylene ammonia can be used for cooling down to 150–50 K. But
such coolers have a large weight, a finite supply of refrigerant, and a
short lifetime (~1 year). Widely used thermoelectric coolers (TECs)
with longer lifetimes can reach only ~175 K. Cryogenic refrigeration
free from vibrations is important for high-performance IR cameras
and night-vision systems, especially in satellite-borne systems. It is

also important for gamma-ray spectroscopy, which can provide the
required energy resolution when the Ge crystals are cooled down
to ~100 K. Ultrastable optical cavities for metrology are another
application of all-solid-state optical cryocoolers.
There are several ways for lowering the final temperature of the
RE-doped sample and increasing the cooling efficiency. For example,
RE ions with lower energy gap between the manifolds involved into
cooling cycles such as Tm3+, Ho3+, and Dy3+ doped in low-phonon
hosts can provide higher cooling efficiency than Yb3+ ions doped
in the same host material. It is found that the increase in the ion
concentration results in an increase in the absorbed pump power,
but the cooperative effects such as reabsorption and energy transfer
can deteriorate the cooling process. To reach some predetermine
temperature of the sample the pump power, ion concentration, and
the size of the sample have to be properly arranged. Purification
of the host material is very important for all samples and can be
considered as a key parameter for the practical realization of optical
coolers for any size range from centimeter to nanometer size.
Molecular species, which usually have vibration modes with higher
energies than the optical phonons of the host, reduced quantum
efficiency of the active RE ions participating in the cooling process.
The TM impurities such as Fe2+, Cu2+, Co2+, and Ni2+ have transitions
with high cross sections in the near-IR. They can deteriorate laser
cooling. The RE impurities such as Er3+, Tm3+, and Ho3+ in the Yb3+-
doped samples are undesirable too, but since the excited states of RE
ions have higher quantum efficiency than the excited states of TM
ions the RE impurities are not so detrimental for the cooling process
as TM impurities and a further decrease of TM impurities is a high
priority problem on the way to high-quality solid-state samples for
laser cooling.
References
1. Hänsch, T. W., Schawlow, A. L. (1975). Cooling of gases with laser

radiation, Opt. Commun., 13, pp. 68–69.
2. Chu, S., Cohen-Tannoudji, C., Philips, W. D. (1997). For development of
methods to cool and trap atoms with laser light, Nobel Prize in Physics,
http://nobelprize.org.
References 75
3. Lenard, P., Schmidt, F., Tomascheck, R. (1928). Handbuch der

Experimentalphysik, 1. Teil Akademische Verlagsgesellschaft, Leipzig.
Lumineszenz- und Temperature-strahlung, Z. Phys., 57, pp. 739–746.
5. Vavilov, S. I. (1945). Some remarks on the Stokes of fluorescent liquids,
J. Phys. USSR, 9, pp. 68–73.
6. Vavilov, S. I. (1946). Photoluminescence and thermodynamics, J. Phys.
USSR, 10, pp. 499–501.
7. Lomonosov, M. V. (1934) Complete papers VI, Leningrad. Ломоносов,
М. В. (1934) Собр. Соч., Ленинград. (in Russian).
8. Landau, L. D. (1946). On the thermodynamics of photoluminescence, J.
Phys. USSR, 10, pp. 503–506.
9. Kastler, A. (1950). Some suggestions concerning the production and
detection by optical means of inequalities in populations of levels of
spatial quantization in atoms, J. Phys. Radium, 11, pp. 255–265.
10. Yatsiv, S. (1961). Anti-Stokes fluorescence as a cooling process,
in Advances in Quantum Electronics (ed. Singer, J. R.) pp. 200–213
(Columbia Univ. Press, New York).
11. Kushida, T., Geusic, J. E. (1968). Optical refrigeration in Nd-doped
yttrium aluminum garnet, Phys. Rev. Lett., 21, pp. 1172–1175.
Stokes scattering, Phys. Rev. Lett., 46, pp. 236–239.
13. Epstein, R. I., Buchwald, M. I., Edwards, B. C., Gosnell, T. R., Mungan, C.
E. (1995). Observation of laser-induced fluorescent cooling of a solid,
Nature, 377, pp. 500–502.
14. Melgaard, S. D, Seletskiy, D. V., Di Lieto, A., Tonelli, M., Sheik-Bahae,
M. (2013). Optical refrigeration to 119 K, below National Institute of
Standards and Technology cryogenic temperature, Opt. Lett., 38, pp.
1588–1590.
15. Soares de Lima Filho, E., Nemova, G., Loranger, S., Kashyap, R. (2013).
Laser-induced cooling of a Yb:YAG crystal in air at atmospheric
pressure, Opt. Express, 21, pp. 24711–24720.
16. Melgaard, S. D., Seletskiy, D. V., Polyak, V., Asmerom, Y., Sheik-Bahae, M.
(2014). Identification of parasitic losses in Yb:YLF and prospects for
optical refrigeration down to 80 K, Opt. Express, 22, pp. 7756–7764.
17. Melgaard, S. D., Seletskiy, D. V., Albrecht, A., Sheik-Bahae, M. (2015).
First solid-state cooling below 100 K, SPIE Newsroom (March 13,
2015). doi:10.1117/2.1201503.005790.
18. Nemova, G., Kashyap, R. (2010). Laser cooling of solids, Rep. Prog.
Phys., 73, pp. 086501-1–20.
19. Zhang, J., Li, D., Chen, R., Xiong, Q. (2013). Laser cooling of a
semiconductor by 40 kelvin, Nature, 493, pp. 504–508.
20. Weinstein, M. A. (1960). Thermodynamic limitation on the conversion
of heat into light, J. Opt. Soc. Am., 50, pp. 597–602.
21. Chukova, Yu. P. (1969). Thermodynamic limit of the luminescence
efficiency, JETP Lett., 10, pp. 294–296.
22. Chukova, Yu. P. (1971). Thermodynamic limit to the efficiency of
broadband photoluminescence, Bull. Acad. Sci. USSR Phys. Ser., 35, p.
1349.
23. Chukova, Yu. P. (1974). Influence of excitation-line characteristics
on efficiency of spectral conversion of energy by ions of trivalent
neodymium in yttrium-aluminum garnet, Bull. Acad. Sci. USSR Phys.
Ser., 38, p. 57.
24. Chukova, Yu. P. (1976). The region of thermodynamic admissibility of
light efficiencies larger than unity, Sov. Phys. JETP, 41, pp. 613–616.
25. Sheik-Bahae, M., Epstein, R. I. (2009). Laser cooling of solids, Laser
Photonics Rev., 3, pp. 67–84.
26. Hehlen, M. P., Sheik-Bahae, M., Epstein, R. I. (2014). Solid-state optical
refrigeration, in Handbook on the Physics and Chemistry of Rare Earths,
Vol. 45, pp. 179–260.
27. Luo, X., Eisaman, M. D., Gosnell, T. R., (1998). Laser cooling of a solid by
21 K starting from room temperature, Opt. Lett., 23, pp. 639–641.
28. Gosnell, T. R. (1999). Laser cooling of a solid by 65 K starting from
room temperature, Opt. Lett., 24, pp. 1041–1043.
29. Kiel, A. (1964). Multi-phonon spontaneous emission in paramagnetic
crystals, in Third International Conference on Quantum Electronics
(Grivet, P., Bloembergen, N., eds.), pp. 765–772 (Columbia University
Press, New York).
30. Riseberg, L. A., Moos, H. W. (1968). Multiphonon orbit-lattice relaxation
of excited states of rare earth ions in crystals, Phys. Rev., 174, pp. 429–
438.
31. Hehlen, M. P. (2009). Design and fabrication of rare-earth doped laser
cooling materials, in Optical Refrigeration: Science and Applications of
Laser Cooling of Solids (Epstein, R. I., Sheik-Bahae, M., eds.), pp. 37–74
(Wiley, Weinheim, Germany).
32. Auzel, F. (2004). Upconversion and anti-Stokes processes with f and d
ions in solids, Chem. Rev., 104, pp. 139–173.
References 77
33. Heeg, B., DeBarber, P. A., Rumbles, G. (2005). Influence of fluorescence

reabsorption and trapping on solid-state optical cooling, Appl. Opt., 44,
pp. 3117–3124.
34. Chandrasekhar, S. (1943). Stochastic problems in physics and
astronomy, Rev. Mod. Phys., 15, pp. 1–89.
35. Olson, R. W., Loring, R. F., Fayer M. D. (1981). Luminescent solar
concentrators and the reabsorption problem, Appl. Opt., 20, pp. 2934–
2940.
36. Goutaudier, C., Lebbou, K., Guyot, Y., Ito, M., Canibano, H., Hassouni, A.
E., Laversenne, L., Cohen-Adad, M. T., Boulon, G. (2003). Advances in
fibre crystals: growth and optimization of spectroscopic properties for
Yb3+-doped laser crystals, Ann. Chim. Mater., 28, pp. 73–88.
37. Förster, Th. (1948). Zwischenmolekulare Energiewanderung und
Fluoreszenz, Ann. Phys., 6, pp. 55–75.
38. Dexter, D. L. (1953). A theory of sensitized luminescence in solids, J.
Chem. Phys., 21, pp. 836–850.
39. Auzel, F., Bonfigli, F., Gagliari, S., Baldacchini, G. (2001). The interplay of
self-trapping and self quenching for resonant transitions in solids; role
of a cavity, J. Lumin., 94–95, pp. 293–297.
40. Jubera, V., Sablayrolles, J., Guillen, F., Decourt, R., Couzi, M., Garcia, A.
(2009). From the infrared to the visible range: spectroscopic studies of
ytterbium doped oxyborates, Opt. Commun., 282, pp. 53–59.
41. Nemova, G., Kashyap, R. (2015). Optimization of optical refrigeration
in Yb3+:YAG samples, J. Lumin., 164, pp. 99–104.
42. Malinowski, M., Kaczkan, M., Piramidowicz, R., Frukacz, Z., Sarnecki,
J. (2001). Cooperative emission in Yb3+:YAG planar epitaxial
waveguides, J. Lumin., 94–95, pp. 29–33.
43. Ghasemkhani, M., Albrecht, A. R., Melgaard, S. D., Seletskiy, D. V.,
Cederberg, J., Sheik-Bahae, M. (2014). Intracavity optical refrigeration
to 130 K using high-power vertical external-cavity surface-emitting
laser (VECSELs), Proc. SPIE, 9000, pp. 900005-1–7.
44. Edwards, B. C., Buchwald, M. I., Epstein, R. I. (1998). Development of
the Los Alamos solid-state optical refrigerators, Rev. Sci. Instrum., 69,
pp. 2050–2055.
45. Edwards, B. C., Anderson, J. E., Epstein, R. I., Mills, G. L., Mord, A. J.
(1999). Demonstration of a solid-state optical cooler: an approach to
cryogenic refrigeration, J. Appl. Phys., 86, pp. 6489–6493.
46. Hoyt, C. W., Hasselbeck, M. P., Sheik-Bahae, M., Epstein, R. I., Greenfield,
S., Thiede, J., Distel, J., Valencia, J. (2003). Advances in laser cooling of
thulium-doped glass, J. Opt. Soc. Am. B, 20, pp. 1066–1075.
47. Seletskiy, D. V., Hasselbeck, M. P., Sheik-Bahae, M. (2010). Resonant
cavity-enhanced absorption for optical refrigeration, Appl. Phys. Lett.,
96, pp. 181106-1–3.
48. Seletskiy, D. V., Hasselbeck, M. P., Sheik-Bahae, M., Epstein, R. I., Bigotta,
S., Tonelli, M. (2008). Cooling of Yb:YLF using cavity enhanced resonant
absorption, Proc. SPIE, 6907, pp. B9070–B9079.
49. Seletskiy, D. V., Melgaard, S. D., Bigotta, S., Di Lieto, A., Tonelli, M.,
Epstein, R. I., Sheik-Bahae, M. (2009). Demonstration of an optical
cryocooler, CLEO/IQEC 2009, post deadline submission.
50. Weber, M. J. (1985). Laser spectroscopy of glasses, Ceram. Bull., 64, p.
1439.
51. MacFarlane, R. M., Shelby, R. M. (1987). Homogeneous line broadening
of optical transitions of ions and molecules in glasses, J. Lumin., 36, pp.
179–207.
52. Yen, W. M., Brundage, R. T. (1987). Fluorescence line narrowing in
inorganic glasses: linewidth measurements, J. Lumin., 36, pp. 209–220.
53. Siegman, A. E. (1986). Lasers (University Science Books. Mill Valley,
CA).
54. Pellegrino, J. M., Yen. W. M., Weber, M. J. (1980). Composition
dependence of Nd3+ homogeneous linewidths in glasses, J. Appl. Phys.,
51, pp. 6332–6336.
55. Lei, G., Anderson, J. E., Buchwald, M. I., Edwards, B. C., Epstein, R. I.
(1998). Determination of spectral linewidths by Voigt profiles in Yb3+-
doped fluorozirconate glasses, Phys. Rev. B, 57, pp. 7673–7678.
56. Hoyt, C. W., Sheik-Bahae, M., Epstein, R. I., Edwards, B. C., Anderson, J.
E. (2000). Observation of anti-Stokes fluorescence cooling in Thulium-
doped glass, Phys. Rev. Lett., 85, pp. 3600–3603.
57. Bowman, S. R., Mungan, C. E. (2000). New materials for optical cooling,
Appl. Phys. B, 71, pp. 807–811.
58. Epstein, R. I., Brown, J. J., Edwards, B. C., Gibbs, A. (2001). Measurements
of optical refrigeration in ytterbium-doped crystals, J. Appl. Phys., 90,
pp. 4815–4819.
59. Poulain, M., Poulain, M., Lucas, J. (1975). Verres fluores au tetrafluorure
de zirconium proprietes optiques d’un verre dope au Nd3+, Mater. Res.
Bull., 10, pp. 243–246.
References 79
60. Tsoukala, V. G., Schoeder, J., Floudas, G. A., Thompson, D. A. (1987).

Intrinsic Rayleigh scattering in fluoride glasses, Mater. Sci. Forum.,
19–20, pp. 637–646.
61. Takahashi, S. (1992). Prospects for ultra-low loss using fluoride glass
optical fiber: a review, J. Non-Cryst. Solids, 140, pp. 172–178.
62. Gan, F. (1995). Optical properties of fluoride glasses: a review, J. Non-
Cryst. Solids, 184, pp. 9–20.
63. Thiede, J., Distel, J., Greenfield, S. R., Epstein, R. I. (2005). Cooling to 208
K by optical refrigeration, Appl. Phys. Lett., 86, pp. 154107-1–2.
64. Murtagh, M. T., Sigel, G. H., Fajardo, J. C., Epstein, R. I. (1999). Laser
induced fluorescent cooling of rare-earth–doped fluoride glasses, J.
Non-Cryst. Solids, 253, pp. 50–57.
65. Fernandez, J., Mendioroz, A., Garcia-Adeva, A. J., Balda, R., Adam, J.
L. (2000). Anti-Stokes laser-induced internal cooling of Yb3+-doped
glasses, Phys. Rev. B, 62, pp. 3213–3217.
66. Fernandez, J., Mendioroz, A., Garcia, A. J., Balda, R., Adam, J. L. (2001).
Laser-induced internal cooling of Yb-doped fluoride-based glasses, J.
Alloys Compd., 323–324, pp. 239–244.
67. Fernandez, J., Mendioroz, A., Garcia, A. J., Balda, R., Adam, J. L.,
Arriandiaga, M. A. (2001). On the origin of anti-Stokes laser-induced
cooling of Yb3+-doped glass, Opt. Mater., 16, pp. 173–179.
68. Fernandez, J., Garcia-Adeva, A. J., Balda, R. (2012). Anti-Stokes laser-
induced cooling in rare-earth doped low phonon materials, Opt. Mater.,
34, pp. 579–590.
69. Fernandez, J., Garcia-Adeva, A. J., Balda, R. (2006). Anti-Stokes laser
cooling in bulk erbium-doped materials, Phys. Rev. Lett., 97, pp.
033001-1–4.
70. Nostrand, M. C., Page, R. H., Payne S. A., Isaenko L. I., Yelisseyev, A. P.
(2001). Optical properties of Dy3+ - and Nd3+-doped KPb2Cl5, J. Opt.
Soc. Am. B, 18, pp. 264–276.
71. Brown, Ei E., Hömmerich, U., Kumi-Barimah, E., Bluiett, A., Trivedi,
S. B. (2015). Comparative spectroscopic studies of Ho: KPb2Cl5, Ho:
KPb2Br5, and Ho: YAG for 2 μm laser cooling applications, Proc. SPIE,
9380, pp. 93800O-1–7.
72. Patterson, W., Hasselbeck, M. P., Sheik-Bahae, M., Bigotta, S., Parisi, D.,
Toncelli, J., Tonelli, M., Epstein, R. I., Thiede, J. (2004). Observation of
optical refrigeration in Tm3+:BaY2F8, Lasers and Electro-Optics (CLEO
(San Francisco, CA, USA).
73. Patterson, W., Bigotta, S., Sheik-Bahae, M., Parisi, D., Tonelli, M., Epstein,
R. I. (2008). Anti-Stokes luminescence cooling of Tm3+doped BaY2F8,
Opt. Express, 16, pp. 1704–1710.
74. Mungan, C. E., Buchwald, M. L., Edwards, B. C., Epstein, R. I., Gosnell,
T. R. (1997a). Laser cooling of a solid by 16 K starting from room-
temperature, Phys. Rev. Lett., 78, pp. 1030–1033.
75. Mungan, C. E., Buchwald, M. I., Edwards, B. C., Epstein, R. I., Gosnell,
T. R., (1997b). Internal laser cooling of Yb3+-doped glass measured
between 100 and 300 K, Appl. Phys. Lett., 71, pp. 1458–1460.
76. Rayner, A., Hirsch, M., Heckenberg, N. R., Rubinsztein-Dunlop, H.
(2001). Distributed laser refrigeration, Appl. Opt., 40, pp. 5423–5429.
77. Heeg, B., Stone, M. D., Khizhnyak, A., Rumbles, G., Mills, G., DeBarber, P.
A. (2004). Experimental demonstration of intracavity solid-state laser
cooling of Yb3+:ZrF4-BaF2-LaF3-AlF3-NaF glass, Phys. Rev. A, 70, pp.
021401-1–4.
78. Seletskiy, D. V., Melgaard, S. D., Epstein, R. I., Di Lieto, A., Tonelli, M.,
Sheik-Bahae, M. (2011). Local laser cooling of Yb:YLF to 110 K, Opt.
Express, 19, pp. 18229–18236.
79. Seletskiy, D. V., Melgaard, S. D., Epstein, R. I., Di Lieto, A., Tonelli,
M., Sheik-Bahae, M. (2012). Precise determination of minimum
achievable temperature for solid-state optical refrigeration, J. Lumin.,
arXiv:1203.5967
80. Nemova, G., Kashyap, R. (2014). Laser cooling in Yb3+:YAG, J. Opt. Soc.
Am. B, 3, pp. 340–348.
81. Bigotta, S., Parisi, D., Bonelli, L., Toncelli, A., Tonelli, M., Di Lieto, A.
(2006). Spectroscopic and laser cooling results on Yb3+-doped BaY2F8
single crystal, J. Appl. Phys., 100, pp. 013109-1–7.
82. Bigotta, S., Parisi, D., Bonelli, L., Toncelli, A., Di Lieto, A., Tonelli, M.
(2006). Laser cooling of Yb3+-doped BaY2F8 single crystal, Opt. Mater.,
28, pp. 1321–1324.
83. Bigotta, S., Di Lieto, A., Parisi, D., Bonelli, L., Toncelli, A., Tonelli,
M. (2007). Single fluoride crystals as materials for laser cooling
application, Proc. SPIE, 6461, pp. 64610E–646110E.
84. Nemova, G., Kashyap, R. (2012). Laser cooling with Tm3+-doped oxy-
fluoride glass ceramic, J. Opt. Soc. Am. B, 29, pp. 3034–3038.
85. Soares de Lima Filho, E., Krishnaiah, K. V., Ledemi, Y., Yu Ye-Jin,
Messaddeq, Y., Nemova, G., Kashyap, R. (2015) Ytterbium-doped glass-
ceramics for optical refrigeration, Opt. Express, 23, pp. 4630–4640.
References 81
86. Boccara, C., Fournier, D., Jackson, W., Amer, N. M. (1980). Sensitive
photothermal deflection technique for measuring absorption in
optically thin media, Opt. Lett., 5, pp. 377–379.
87. Jackson, W. B., Amer, N. M., Boccara, A. C., Fournier, D. (1981).
Photothermal deflection spectroscopy and detection, Appl. Opt., 20,
pp. 1333–1344.
88. Garcia-Adeva, A. J., Balda, R., Al Saleh M., Garcia-Revilla, S., Fernandez,
J. (2011). Local internal and bulk optical cooling in Nd-doped crystals
and nanocrystalline powders revisited, Proc. SPIE, 7951, pp. 795106-
1–10.
89. Garcia-Adeva, A. J., Balda, R., Al Saleh, M., Garcia-Revilla, S., Sola, D.,
Fernandez, J. (2012). Optical cooling of Nd-doped solids, Proc. SPIE,
8275, pp. 827502-1–14.
90. Ghasemkhani, M., Seletskiy, D. V., Sheik-Bahae, M. (2011). Sensitive
thermal reflectance measurement for laser cooling applications, Proc.
SPIE, 7951, pp. 79510J-1–6.
91. Ghasemkhani, M., Seletskiy, D. V., Sheik-Bahae, M. (2012). Polarization-
resolved optical metrology for noncontact thermometry, Proc. SPIE,
8275, pp. 82750J-1–5.
92. Farley, C. III, Reddy, B. R. (2011). Mach-Zehnder interferometric
measurement of laser heating/cooling in Yb3+:YAG, Proc. SPIE, 7951,
pp. 79510I-1–8.
93. Rayner. A., Friese, M. E. J., Truscott, A. G., Heckenberg, N. R., Rubinsztein-
Dunlop, H. (2001). Laser cooling of a solid from ambient temperature,
J. Mod. Opt., 48, pp. 103–114.
94. Condon, N. J., Bowman, S. R., O`Connor, P. O., Quimby, R. S., Mungan, C.
E. (2009). Optical cooling in Er3+:KPb2Cl5, Opt. Express, 17, pp. 5466–
5472.
95. Seletskiy, D. V., Melgaard, S. D., Bigotta, S., Di Lieto, A., Tonelli, M., Sheik-
Bahae, M. (2010). Laser cooling of solids to cryogenic temperatures,
Nat. Photonics, 4, pp. 161–164.
Chapter 3
Progress toward Laser Cooling of

Thulium-Doped Fibers
Nai-Hang Kwong,a Rolf Binder,a Dan Nguyen,b and

Arturo Chavez-Pirsonb
aCollege
of Optical Sciences, University of Arizona, Tucson, AZ 85721, USA
bNPPhotonics, 9030 Rita Road, Tucson, AZ 85747, USA
kwong@optics.arizona.edu
3.1 Introduction
Laser cooling has been successfully demonstrated on rare earth–
doped glass [1–3] and CdS semiconductor crystals [4]. (For recent
general reviews of progress in solid laser cooling, see, for example,
Refs. [5, 6].) These advances prompted considerations on how
laser cooling of solids may be implemented on a practical scale
for applications. In particular, it was noted (2010 Office of Naval
Research STTR Solicitation and private communications with Dr.
Deborah Van Vechten, ONR) that the possibility of transferring
waste heat as photons to remote locations is a desirable advantage
of optical refrigeration. These considerations have motivated us to

www.panstanford.com
84 Progress toward Laser Cooling of Thulium-Doped Fibers
examine the feasibility of the laser cooling of doped optical fibers.

In this chapter, we discuss the progress in our theoretical and
experimental efforts in this direction.
An all-fiber cooling approach, if successful, has many practical
advantages. A fiber device is typically lightweight, small-sized,
vibrationless, and almost maintenance free. Fibers have long
operation lifetimes, even under critical (e.g., in outer space)
environmental conditions. The doping rare earth atoms have been
demonstrated to yield high laser cooling efficiency in small glass
host samples. It is also convenient to integrate multiple fiber lasers
and fiber coolers in one cooling configuration and to design devices
to transfer waste (heat) photons out of the fiber cooler to desired
locations. Our experimental cooling fibers are made of thulium-
doped germinate or tellurite glass which has a potentially high
cooling efficiency. A sketch of a design of a fiber cooler is shown in
Fig. 3.1.
Tm-doped
cooling fiber
Heating
Tm-doped
fiber laser PL Source PF
Heat dump
Figure 3.1 Schematic of the fiber cooling geometry. A fiber laser pumps
the fiber cooling segment, which is attached to a heat source.
The up-converted photoluminescence light can, in principle, be
piped arbitrary distances before being dumped and converted
to heat.
Expectedly there are challenging issues to be tackled. For example,

the purity level of the (undoped) fiber needs to be high to avoid
excessive parasitic absorption of photons by impurities. When the
fiber cooler and the object to be cooled are macroscopic (centimeter
scale or larger), managing the (conductive and convective) heat and
photon flows is also a pertinent issue. These and related issues will
be discussed below.
In Section 3.2, we describe our theoretical developments in
this investigation. We discuss the calculation of the cooling powers
Theoretical Developments 85
of the Tm atoms and of a model isolated fiber and the simulation

of heat flow in the cooling of the heat load. In Section 3.3, we
discuss experimental progress in fabricating high-purity fiber and
temperature measurements under optical pumping conditions.
3.2 Theoretical Developments
3.2.1 Modeling the Cooling of Thulium Atoms

The basic cooling agents in our setup are the thulium (Tm) atoms
doping the fiber. We describe in this subsection a photoexcitation
model of a Tm atom in a glass host, which we use to estimate its laser
cooling efficiency. More detail can be found in [7]. The model, given
in Ref. [8], is shown in Fig. 3.2. It consists of four energy levels, each
with a degree of degeneracy gi, as designated in the figure. The lower
two levels represent the manifold of states created by the splitting of
the 13-fold degenerate ground state (3H6) of Tm by interactions with
the glass host, and the upper two levels represent the split ninefold
excited-state (3F4) manifold. The pump laser energy hn is tuned
to the energy between |1> and |2>, which is the lowest excitation
energy between the two manifolds. The pumped Tm atoms de-
excite (incoherently) to a level in the ground-state manifold via
luminescence, at rate Wrad, and nonradiative transitions, at a rate
Wnr. If the de-excitation happens sufficiently slowly, the excited
atoms have time to thermalize in the excited-state manifold. The
luminescent photons are thus at higher energies than that of the
pump photons. In each cycle of operation in which a Tm atom
absorbs a pump photon and emits a luminescent photon, which then
escapes, heat is removed from the system.
The population dynamics of the four-level model obeys the
following rate equations [8]:
dN1 Ê g ˆI R Ê g -d E /k T ˆ
= -s 12 Á N 1 - 1 N2 ˜ inc + (N2 + N3 ) - w1 Á N1 - 1 N0e g B ˜
dt Ë g2 ¯ hn 2 Ë g0 ¯
(3.1)
dN 2  g I  g 
= s 12  N1 − 1 N 2  inc − RN 2 + w2  N 3 − 3 N 2 e − d Eu /kBT 
dt  g2  hn  g2 
(3.2)
dN3 Ê g ˆ
= -RN3 - w2 Á N3 - 3 N2e -d Eu /kBT ˜ (3.3)
dt Ë g2 ¯
Ntotal = N0 + N1 + N2 + N3 (3.4)
where Ni and gi are the population density and degree of degeneracy,

respectively, of level |i>, and Ntotal is the total density of Tm atoms.
The pump laser has intensity Iinc and frequency n. R = 2Wrad + 2Wnr,
and s12 denotes the photoabsorption cross section of the Tm atoms.
dEg and w1 are the energy difference between the two levels and
the intramanifold relaxation rate in the ground-state manifold, and
dEu and w2 are the corresponding quantities for the excited-state
manifold. T is the sample temperature, and kB is the Boltzmann
constant. In Eq. 3.1, the first term on the right side represents
stimulated absorption from and emission into the pump laser
mode. The second term models the decay of the excited manifold
population and the third term the thermalization within a manifold.
The terms in Eqs. 3.2 and 3.3 have the same interpretation.
|3> g3 = 4
δEu w2
|2> g2 = 5
hν Wrad Wnr
|1> g1 = 6
δEg w1
|0> g0 = 7
Figure 3.2 Sketch of the four-level model used to simulate cooling of Tm

(after Ref. [8]). The ground-state manifold represents the 3H6
state, and the excited-stated manifold 3F4. Reprinted with
permission from Ref. [7], Copyright 2011, Society of Photo
Optical Instrumentation Engineers.
From the solution of the rate equations, we can obtain the

absorbed power density as
Ê gN ˆ
Pabs = s 12 N1 Á 1 - 1 2 ˜ Iinc + a Iinc (3.5)
Ë g2 N1 ¯
and the luminescence power density as
Prad = Wrad[N2(E21 + E20) + N3(E31 + E30)] (3.6)
where Eij = Ei – Ej and Ei is the energy of level |i>. In Eq. 3.5, apart from
the power absorbed by the Tm atoms, we have introduced a parasitic
absorption contribution aIinc, which takes into account the photon
power absorbed by other excitation modes, such as (unwanted)
impurities in the glass host, and converted into heat. 1/a represents
the spatial scale of this absorption. Assuming all luminescence
photons to be able to leave the system, the cooling performance of
the Tm-doped glass is characterized by the net power it gains
Pnet = Pads – Prad (3.7)
and its cooling efficiency
ec = –Pnet/Pabs (3.8)
A negative Pnet signifies that cooling is possible.
Figure 3.3 shows some calculated results for the cooling
power Pnet as a function of the (temporally constant) laser pump
intensity based on the steady-state solutions of Eqs. 3.1–3.4. The
different curves are calculated with different values of the parasitic
absorption rate a. The other parameters are set at pump wavelength
= 1.94 mm (hn = 0.639 eV = 1.024 × 10–19 J), E21 = hn , s12 = 2 × 10–26
m2, Wrad = 83 s–1, Wnr = 0.64 s–1, dEu = 1.15 × 10–20 J, dEg = 1.00 ×
10–20 J, w1 = w2 = 1000 s–1, Ntotal = 3.43 × 1026 m–3, and T = 300 K.
These parameter values are mostly taken from Ref. [9], where they
are validated in an experimental study of laser cooling of Tm-doped
ZBLAN glass. Although our glass host is germanate or tellurite, the
adopted parameter values can be considered adequate in getting a
reasonable estimate of the cooling power.
The trend of the curves in Fig. 3.3 shows the detrimental effect
of the parasitic absorption: cooling is not possible for a larger than
a value of around 2 × 10–2 m–1. In each curve, the deterioration of
the cooling power at high pump intensity is due to the saturation
of the absorption/luminescence of the Tm atoms. For steady-state
solutions with no parasitic absorption (a = 0), the cooling power Pnet

can be shown to approach, as Iinc tends to infinity, the limit
Ê hn ˆ
lim Pnet (a = 0) = Rhn(1 + f32 )Á 1 - hq m ˜ N2 (3.9)
Iinc Æ• Ë hn ¯
where
d Eg f32 Wrad
hn m = hn + + d E u , hq = (3.10)
2 1 + f32 Wrad + Wnr
Ê w2 ˆ g3 -d Eu /kBT
f32 = Á e (3.11)
Ë w2 + R ˜¯ g2
and the density N2 approaches the limit
-1
N2 Èg Ê g ˆ Ê R ˆ˘ g1 -d Eg /kBT
lim =Í 1Á + Á 1 + (1 + f32 ) ˙ , g= e
Iinc Æ• N total ˜
ÍÎ g2 Ë g + 1 ¯ Ë 2 gw1 ˜¯ ˙˚ g0
(3.12)
Optimal pump power* (W)

0 0.98 2 2.9 3.9
4.3×10-4
10
a(10-3m-1)
2.2×10-4
5 a(dB/cm)
Pnet (W/cm3)
0.0
0
0 1 2
Optimal pump irradiance (109 W/m2)
Pump Irradiance (109 W/m2)
Figure 3.3 Calculated cooling power (Pnet < 0 signifies cooling) of

the fiber versus pump irradiance for a range of values of
parasitic absorption rate a. The dotted line and the inset
show the optimal cooling power for each a. For each curve,
the deterioration of cooling power at high irradiance is due to
absorption saturation of the thulium ions.
In Eq. 3.10, hnm is defined as the mean luminescence energy and

hq is the quantum efficiency. Equation 3.9 implies that a necessary
condition for cooling to occur is
hn m
hq > 1 (3.13)
hn
The saturation of the cooling capacity of the Tm atoms at
high pump intensities, as implied by Eq. 3.9, allows the parasitic
absorption to eventually dominate and disable the cooling of
the fiber. The optimal cooling power for each a is marked on the
corresponding curve in Fig. 3.3 and plotted in the inset of the figure.
We have also repeated the cooling power calculation using a more
elaborate model with 13 levels in the ground manifold and 9 levels
in the excited manifold, obtaining similar results as in Fig. 3.3.
3.2.2 Cooling Power of a Pumped, Doped Fiber in Free

Space
In this subsection, we describe some theory developments in
support of our effort to estimate the temperature and infer the
cooling power of the irradiated, isolated fiber from its measured
emission spectrum. After the fiber is irradiated at low intensity,
its emission spectrum shows a systematic dependence on the
temperature at which it is set. This dependence allows the emission
spectrum to be calibrated for temperature measurements. Then in
a laser cooling measurement, when the fiber is pumped at higher
intensities, its emission spectrum can be measured, and the fiber’s
temperature can be determined by comparing this spectrum to the
calibration spectra. This strategy was successfully used in Refs. [10,
11]. We have applied it to estimate the temperature of the pumped
Tm-doped fiber. The experimental measurement of the spectra
is discussed in Section 3.3. The theoretical aspects are treated in
this subsection. We show that the temperature dependence of
the emission spectrum can be reasonably described by a relation
between the absorption and emission spectra. With the estimated
temperature of the laser-pumped fiber, we infer from it the fiber’s
cooling power by performing heat flow simulations in the fiber.
The obtained cooling power is compared to the predictions of the
microscopic model in the previous subsection.
Figure 3.4 shows a drawing of the cross section of one of the

fibers we used in this study. It consists of a core fiber of Tm-doped
tellurite glass of radius 52.5 mm surrounded by a cladding with an
outer radius of the 62.5 mm. We seek to understand the temperature
variations of the fiber’s emission spectrum by relating it through
general thermal arguments to the photoabsorption spectrum [12].
Photoabsorption and emission are governed by the same quantum
transition matrix elements between the levels in the ground (3H6)
manifold and those in the excited (3F4) manifold of the Tm ions.
We assume that the excited Tm atoms have thermalized among the
states in the excited manifold before they luminesce. Then their
emission spectrum at temperature T, L(l, T), is given by [13]
hc Ê 1 ˆ
C1 - -C
kBT ËÁ l 2 ¯˜ (3.14)
L( l ,T ) = s ( l ,T ) e
l2
where l denotes the wavelength, c is the speed of light in vacuum,
and s(l,T) is the single-ion absorption cross section averaged
over an ensemble of ions thermally distributed at temperature T
within the ground manifold. C1 and C2 are parameters that depend
on the material but not on T or l. Knowledge of C1 and C2 is not
necessary if we are interested only in the spectral distribution but
not in the absolute magnitude of L(l, T). If the absorption cross
section of the sample is independently measured, then Eq. 3.14
provides a consistency check on the measured emission spectra
and can indicate, for example, whether the excited ions have
thermalized in the excited manifold. Equation 3.14 differs from two
commonly used absorption/emission relations: the Roosbroeck–
Shockley relation [14] and the Kubo–Martin–Schwinger (KMS)
relation [15] in the conditions under which they are applicable.
For the Roosbroeck–Shockley relation to apply, both emission and
absorption measurements are performed on a system at thermal
equilibrium with the surroundings (the photon bath). In our setup
here, where Eq. 3.14 applies, the emission is measured from excited
ions that are constrained to thermalize only within the excited (3F4)
manifold and so are not in (complete) thermal equilibrium with the
surroundings. The KMS relation has been used in studies of laser
cooling of semiconductors [16]. In this context, both absorption and
emission are to be measured on a semiconductor sample that is in
a metastable quasi-equilibrium state with a fixed density of excited

electron–hole pairs.
Figure 3.4 Schematic drawing of the cross section of the fiber. The core
is a Tm-doped tellurite fiber of radius 52.5 mm, and the outer
radius of the cladding is 62.5 mm.
We have measured the absorption cross section of Tm ions in a

germinate glass sample at T = 300 K, which is plotted as the black
curve in the left panel of Fig. 3.5. Using this cross section, we have
calculated through Eq. 3.14 the temperature-dependent emission
spectrum. The (scaled) emission spectrum at 300 K is plotted as the
red curve in the graph with the absorption cross section. The spectra
for a range of temperature are plotted in the right panel of Fig. 3.5,
which are to be compared with the experimental calibration spectra.
The general trend of the theoretical emission spectra is matched by
that in the experiment. Despite the difference in glass hosts in the
absorption (germanate) and emission (tellurite) measurements, the
spread of the theoretical spectra over 20 K is consistent with that in
our experimental calibration spectra (see Section 3.3).
The measured emission spectra of the irradiated fiber give the
temperature as a function of the pump laser intensity. We obtain the
fiber’s cooling power from the temperature by performing heat flow
simulations in a thermal model of the fiber. In this model the fiber
is infinitely long and has a cross section, as shown in Fig. 3.4, with
the dimensions and compositions, as stated there. Heat flows inside
the fiber and cladding by diffusion, and the cladding is thermally

coupled to the outside by thermal radiation exchange across its
outer surface. For each of a range of cooling power densities, we have
numerically evolved the (time-dependent) heat diffusion equation
inside the fiber until the temperature profile reaches a steady state.
0.30 1.0
Absorption (black) [10-20cm2]
0.25 0.9
Emission (red) [arb. units]
0.8
Normalized intensity
0.20 0.7
0.6 285 K
0.15
0.5 290 K
0.4 295 K
0.10
0.15 300 K
0.3 305 K
0.05 0.2 310 K
0.1 315 K
0.00 0.0
1400 1500 1600 1700 1800 1900 2000 1600 1640 1680 17201760 1800 1840 1880 1920
Wavelength [nm] Wavelength (nm)
Figure 3.5 Left panel: the measured photoabsorption cross section of the
Tm-doped glass (black) and the calculated emission spectrum
(red) at 300 K of the pumped fiber according to Eq. 3.14. The
absorption cross section has been smoothed before being used
in Eq. 3.14. Right panel: the calculated emission spectra for
various temperatures. All the curves are normalized to 1 as
their peak values. Reprinted with permission from Ref. [12],
Copyright 2013, Society of Photo Optical Instrumentation
Engineers.
Temperature drop (T–T0) [K]
0
0
Cooling power [W/cm3]
–10 –2
–20 –4
–30
–6
–40
–8
0 –2 –4 –6 –8 0.0 0.1 0.2 0.3 0.4
Cooling power Pnet [W/cm3] Pump Irradiance [109 W/cm3]
Figure 3.6 Left panel: calculated temperature drop of the fiber as a function
of its cooling power Pnet (T0 = 300 K). Right panel: cooling
power of the pumped fiber as extracted from the experimental
emission spectra (black squares) and as calculated by the
four-level model in Section 3.2.1 (red curve). Reprinted with
The boundary condition of radiative heat exchange with the outside

is applied at the outer cladding surface. The outside temperature T0
and the initial temperature of the fiber-cladding structure are set
at 300 K. The calculated steady-state temperature profiles turn out
to be essentially uniform across the structure, making it possible to
characterize the structure’s thermal state by one temperature. The
calculated temperature change induced by switching on the cooling
power density is plotted as a function of the latter in the left panel
of Fig. 3.6. By considering the heat balance across the outer cladding
surface, one can also obtain the following analytic estimate of the
steady-state temperature T of the structure
rcore
2
T 4 = T04 + P (3.15)
2Res net
where rcore is the core radius, R is the outer cladding radius, e is the
emissivity of the cladding surface, and s is the Stefan–Boltzmann
constant. For the parameter values used here, Eq. 3.15 is a good
approximation to the simulation results shown in Fig. 3.6.
Through the relation between temperature and cooling power
in (the left panel of) Fig. 3.6, we have converted the experimental
temperature-pump-intensity relation into one between the cooling
power of the fiber and the pump intensity. In the right panel of Fig.
3.6, this inferred cooling power density (solid squares) is plotted
against the pump laser intensity. For comparison, we have also
plotted the cooling power (red curve), as calculated in the four-level
Tm atom model (see Fig. 3.2), adjusting the parameters to obtain a
good fit to the measured data. While most parameters are assigned
the values used in Fig. 3.3, we have found that the best fit is obtained
by assuming negligible nonradiative decays parasitic absorption,
which is unlikely to be the case for our fiber. Even with this and
other uncertainties, the comparison shows the microscopic model
predictions and the measured data to be qualitatively (to 1 order of
magnitude) consistent.
3.2.3 Heat Transport in a Structure Cooled by a Laser-

Cooled Fiber
To better understand the prospects of the laser-cooled fiber as a
cooler, we have studied possible cooling actions of the fiber on a
material structure in contact with it. In this section, we describe our

model simulations of heat flow in this cooling device.
3.2.3.1 Model of the cooling device

Figure 3.7 shows a schematic drawing of the heat transport simula-
tion model of the cooling device. The cooling fiber, composed of an
active doped inner core and an outer cladding, rests in a V-shaped
groove, which is set in a silica glass base. This structure is capped
by a metallic heat spreader, which is made up of a nickel-chromium
layer on top of a silver layer. The silver reflects the upward-going
fluorescent photons from the cooling fiber. A thermistor and heat-
load resisters are located on the top of the device. The doped fiber
core acts as a heat sink, its cooling power being set as an input to
the simulation. The model accounts for two heat sources within the
simulation domain: power dissipated by the thermistor and heat-
load resistors, and absorption of scattered fluorescence by the metal
heat spreader.
Thermistor, resistors
Metal heat 0.5 mm
spreader
V-groove 0.5 mm
Fiber
3 cm
Silica glass base
y
x Doped core
z Cladding
Fiber cross section
Figure 3.7 3D simulation domain (relative sizes of components not drawn

to scale) of the cooling device. Composition details: the fiber
is made up of an inner doped core (diameter = 50 μm) and an
outer cladding (diameter = 125 μm); the heat spreader is a 100
μm thick silver layer topped by a 50 μm thick nickel-chromium
layer; a high conductivity epoxy forms a 20 μm adhesive layer
and fills the V-groove around the fiber. The inset (lower right
of the figure) shows schematically the fiber cross section with
doped core and cladding.
The simulation domain is the interior of the device. The evolution

of the temperature field T(r), r = (x, y, z), is governed by the heat
diffusion equation with sources
∂T 1 1
= — ◊ (k—T ) + S (3.16)
∂t rc v rc v
where r is the mass density, cn is the heat capacity, and k is the

thermal conductivity. S(r) represents the heat sources/sinks inside
the simulation domain mentioned above. The boundary condition
at the outer surfaces of the cooling device sets the heat flux at the
surface J as equal to the radiative exchange with the outside
(
J = -k—T = s T - TA n
4 4
) (3.17)
where e is the local surface emissivity, s is the Stefan–Boltzmann
constant, n is the local unit normal vector, and TA is the outside
temperature.
3.2.3.2 Numerical simulation of heat flow in the cooling

device model
We have developed a numerical code to solve the heat diffusion Eq.
3.16 in the 3D geometry of the cooling device. The spatial derivatives
are discretized by finite differences. The time marching combines a
2D alternating direction implicit (ADI) method [17] for updating the
solution across the transverse (x, y) plane and a predictor corrector
(PC) method for updating along the longitudinal (z) direction. Since
the 3D simulations incur high computational costs, computational
efficiency has become an essential requirement in the code’s
development.
In Figs. 3.8–3.10, we show some simulation results from a sample
run of the code. In this simulation, the domain is discretized into 200
cells in each of the two transverse directions (x and y) and 10 cells in
the longitudinal (z) direction. The time evolution of the temperature
at the thermistor is plotted in Fig. 3.8. An initial steady state of the
device is set up with the thermistor (heating power = 1 mW) turned
on but the pump laser off. In this state, the heat introduced into the
device by the heat load (thermistor) is balanced by the heat radiated
out at the device’s surfaces. The pump is turned on at t = 0. It has two
effects on the thermal budget of the device: (i) It supplies the doped
fiber core with a cooling power of 8 W/cm3, and (ii) it generates
fluorescent and scattered photons, some of which are absorbed and

converted into heat by the metal spreader layer (see Fig. 3.7). As
shown in Fig. 3.8, the net effect of all the above processes is that the
device is cooled by 1.5 K, reaching a new steady state after several
minutes.
Pump turned on
Core temperature (K)
Cooling by fiber
Heating by thermistor (load)
Time after pump activation (minutes)
Figure 3.8 Time evolution of the temperature at the thermistor, showing

cooling by the laser-pumped fiber. An initial steady state is set
up with the thermistor on and the pump laser off. The pump is
turned on at t = 0, and the temperature drops by 1.5 K several
minutes thereafter.
Figure 3.9 shows the temperature profile across a transverse

cross section (x-y plane) of the device at the mid-point in the
longitudinal (z) direction, at the steady state reached after the pump
is turned on. Figure 3.10 shows the temperature profile across a
vertical cross section (y-z plane) at the midpoint in the x direction
at the same time. The two temperature profiles indicate the flow
of heat from the thermistor, the principal heat source, to the fiber
core, where the heat is absorbed, and to the boundaries of the
device, where it is emitted as thermal radiation. We note that the
temperature drop between the middle and the end surfaces in the z
direction is relatively large (~0.4 K) compared to that between the
top and the bottom (<0.1 K).
Equilibrium after pump activation

500 307.415
450 307.41
400 307.405
350 307.4
300 307.395
Y (microns)
250 307.39
200 307.385
150 307.38
100 307.375
50 307.37
0 307.365
0 50 100 150 200 250 300 350 400 450 500
X (microns)
Figure 3.9 Calculated temperature profile in a transverse cross section

(x-y plane) of the device at the midpoint in the longitudinal
(z) direction, when the system reaches a steady state after the
pump is turned on. The profile indicates the heat flow from the
thermistor (top middle) to the cooling fiber (centered at x =
250 μm, y = 275 μm), where the heat is absorbed, and to the
device boundaries, where it is radiated. The temperature scale
(in K) for the color coding is shown on the right.
Equilibrium after pump activation

500 307.45
450 307.4
400 307.35
350 307.3
Y (microns)
300
307.25
250
307.2
200
250
150 307.15
100 307.1
50 307.05
0 307
0.5 1 1.5 2 2.5
Z (cm)
Figure 3.10 Calculated temperature profile in a vertical (y-z plane) cross

section at the midpoint in the x direction. Note the much larger
temperature drop between the middle and the sides (~0.4 K)
than between the top and the bottom (<0.1 K).
We note that, because of the high computational costs, we have

limited our simulations here to a model domain with transverse
dimensions of 0.5 mm × 0.5 mm (see Fig. 3.7), which is much smaller
than the experimental device. We have also performed simulations
with a 2D version of the code, which essentially assumes the device to
be translationally invariant in the longitudinal (z) direction. Results
of these 2D simulations, and a discussion of the photon reflection
and absorption by the metal heat spreader, can be found in Ref. [18].
3.3 Experimental Developments
3.3.1 Fabrication of Highly Purified Tm-Doped Tellurite

Glasses
Transition metal (TM) impurities in glasses, especially Fe+2 and Cu+2
impurities, pose serious challenges to the cooling processes. The
TMs Fe+2 and Cu+2 have strong and broad absorption bands from
1 to 2 mm (peak absorption is ~1 mm). There are two processes
caused by the impurities that are negative to the cooling processes
in Tm-doped glass: (i) TMs can directly absorb light (both pump
and luminescence) and generate heat, and (2) excited Tm ions
can transfer energy to surrounding impurities, resulting not only
in lower quantum efficiency but also heat generation. We have
fabricated Tm-doped tellurite glasses using raw materials having
two different purified levels. The first Tm-doped tellurite glasses
(TE-I) were fabricated with average Fe+2 and Cu+2 concentrations
of about 3.4 ppm and 0.83 ppm, respectively. The second Tm-doped
tellurite glasses (TE-II) were fabricated using raw materials with
1 to 2 orders higher purification levels. As a result, TM impurities
in glasses TE-II are much lower compared with glasses TE-I.
Concentrations of all TM elements are less than 1 ppm in glasses TE-
II, as shown in Table 3.1.
It is worthwhile to stress that concentrations of all metal
impurities in our Tm+3-doped tellurite glass TE-II are lower than 1
ppm, as shown in Table 3.1. The metal impurities have been reduced
significantly from standard tellurite glasses, in particular Fe2+, and
Cu2+ has been reduced to ~0.6 ppm and ~0.2 ppm, respectively. As
Experimental Developments 99
can be seen from Fig. 3.11, the total loss due to metal impurities
absorption in the fiber glass is about 0.002 dB/cm around 1.9 mm
spectral region where optical cooling processes occur in Tm-doped
glass.
Table 3.1 Metal impurities in Tm+3-doped tellurite glass TE-II
Metals ppm Metals ppm Metals ppm Metals ppm
Ca44(MR) <1 Er167(MR) <0.05 Ni58(MR) 0.77 U238(LR) <0.05
Cd111(LR) 0.16 Eu151(LR) <0.05 Pb208(MR) 0.12 V51(MR) <0.05
Co59(MR) 0.19 Fe56(MR) 0.63 Pd105(LR) <0.05 W182(LR) <0.05
Cr52(MR) 0.08 Ga69(MR) <0.05 Pt195(LR) <0.5 Y89(LR) <0.05
Cs133(LR) <0.05 Ge72(MR) <0.05 Rb85(LR) <0.2 Yb172(LR) <0.05
Cu63(MR) 0.18 Hf178(LR) <0.05 Re187(LR) <0.05 Zn64(MR) <0.2

Source: Ref. [12]
0.020
0.018
Impurities Absorption (dB/cm)
0.016
0.014
0.012
0.010 1ppm Cu+2 abs
0.008
loss in glass
0.006
0.004
1ppm Fe+2 abs
0.002
1ppm Co+2 abs
0.000
0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1
Wavelength (micron)
Figure 3.11 Loss due to metal impurities in Tm+3-doped tellurite glass:
1 ppm Cu, 1 ppm Fe, 1 ppm Co, and total loss due to actual
impurities in glass. Reprinted with permission from Ref. [12],
Engineers.
As is well known, the hydroxyl group –OH also has big impact
on optical cooling due to the following reasons: (i) Energy transfer
from Tm ions in excited levels to –OH proceeds via a nonradiative
decay, which generates heat in the system, and (ii) contributions to
the fiber loss in general (absorption and scattering, for example)
can also generate heat. In this work, we have achieved reducing –OH
absorption in glasses TE-II from 75–90 dB/m normal tellurite glass
to ~8 dB/m at 3.3 mm, as shown in Fig. 3.12.
100
90 TWD-S4
TD-S23
80
TD-S24
OH Absorption (dB/m)
70 TD-S25
TD-S26
60 TD-S27
50
40
30
20
10
0
2250 2500 2750 3000 3250 3500 3750 4000 4250 4500
Wavelength (nm)
Figure 3.12 –OH absorption in Tm-doped tellurite glasses TD-S23 to TD-
S27 compared to standard glass; TDS-25 glass (blue) has been
used for fabrication of cooling fiber Tm-CF2. Reprinted with
3.3.2 Design and Fabrication of Cooling Fibers

The cooling fiber design requires some special properties that
are different compared to fiber lasers and fiber amplifiers. First,
the fiber has features that aim at minimizing the reabsorption of
the fluorescence, which generates heat in the fiber core. Since the
cooling efficiency is on the order of a few percentage points, even
1% of fluorescence reabsorption can have a large negative impact
on the overall cooling efficiency. At the same time, the cooling fiber
configuration should be able to guide the fluorescence light in the
inner cladding. Second, the fiber configuration must satisfy the
thermal conditions that can maximize the transfer of heat from the
heating sample to the fiber core where cooling effect takes place.
We have fabricated three Tm-doped tellurite glass cooling fibers.
All fibers have core glasses doped with 2 wt% Tm concentration.
The main differences between those fibers are the glasses with the
different levels of TM impurities and –OH absorption. To improve
the potential cooling performance in the fibers, the different fiber
sizes are also designed and fabricated to improve pump absorption
and also coupling losses with passive fibers. The specifications of the
first cooling fiber (Tm-CF1) are: 2% Tm-doped tellurite glass (TE-
I); core/clad size (diameter): 50/125 mm. The glass TE-I has been
fabricated with raw materials having a standard level of impurities.
However, the –OH absorption is reduced to ~8 dB/m compared to
75–90 dB/m of standard glasses. The second cooling fiber (Tm-CF2)
is 2% Tm-doped tellurite glass (TE-II); core/clad size: 105/125 mm.
The glass TE-II has been fabricated with highly purified materials
having low impurities (all TMs are less than 1 ppm). Figure 3.13
shows the two cooling fibers Tm-CF1 and Tm-CF-2.
Figure 3.13 Images of two Tm-doped tellurite glass cooling fibers Tm-CF1
(left) and Tm-CF2 (right). Reprinted with permission from Ref.
[12], Copyright 2013, Society of Photo Optical Instrumentation
Engineers.
Using the data of impurities measured in the glasses and

absorption cross sections from Ref. [19] we can calculate the loss
due to impurity absorption. Figure 3.14 shows loss due to impurity

absorption in Tm-CF1 and Tm-CF2.
0.036
0.032
Absorption dB/cm
0.028
0.024
0.020
Tm-CF1
0.016
0.012
0.008
Tm-CF2
0.004
0.000
1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2
Wavelength (micron)
Figure 3.14 Impurities loss in cooling fibers Tm-CF1 and Tm-CF2.
Characterizations of the cooling fibers Tm-CF1 and Tm-CF2

show the total loss of Tm-CF1 and Tm-CF2 at a test wavelength 1310
nm are 0.042 dB/cm and 0.0409 dB/cm, respectively. At first glance,
the total losses in the two fibers are similar. However, the core
diameter of Tm-CF2 is twice that of Tm-CF1, and as a result CF2 has
higher scattering loss due to roughness of the core/clad boundary.
As shown in Fig. 3.14, the loss due to impurities that has a negative
impact on the optical cooling in the second cooling fiber Tm-CF2 has
been significantly reduced to about 0.002 dB/cm from ~0.009 dB/
cm in the first cooling fiber Tm-CF1.
3.3.3 Experimental Demonstration of Optical Cooling in

Fibers
3.3.3.1 Thermistor measurements

Figure 3.15 shows the sample of the experimental setup with
a 3 cm cooling fiber and a thin silver layer that plays the role of a
heat spreader for the heat source. The cooling fiber is then put
in a V-groove holder with a high conductivity epoxy around it.

The epoxy used in the setup is EPO-TEK H20-HC, which is one of
the highest thermal-conductivity (9.96 W/mK) epoxies. A high-
precision NTC MF51E thermistor (by Cantherm) for extremely
accurate temperature measurement is attached to the metal plate.
The extremely small size allows the thermistor to respond quickly to
tiny changes in temperature.
Thermistor Metal plate V-groove glass fiber coupled

to fiber laser
Heaters(resistors)
Coupling fiber
Figure 3.15 The experiment setup and some important parts in the cooling
experiment. The cooling fiber is underneath of the metal
(silver) plate.
The cooling experiments have been carried out in a vacuum

chamber. The experimental setup is shown in Fig. 3.15, including
3 cm of cooling fiber, where one fiber end is fusion-spliced with a
pump-coupling fiber of size 25/250 mm that couples to the fiber laser.
The other end is fusion-spliced to another coupling fiber of the size
105/125 mm, in which the waste photons can be piped substantial
distances and expelled wherever the system designer wishes. The
total loss of the chain is 0.31 dB (loss of two coupling + background
loss in cooling fiber). Experimental data of temperature versus time
in our cooling system in the vacuum chamber are shown in Fig. 3.16.
Note that the potential cooling effect depends on pumping power:
there is little heat removal if the pump power is too low, but the
cooling process could be overwhelmed by heat if the pump power is
too high. Therefore we first measured temperature versus time with
unchanged pump power in ~1 hour, then increasing power to higher
levels. The aim is to find the best range of power for the potential
cooling processes in the fiber.
The experiments are repeated with each pump power after about
5 hours. In the experiment, the initial temperature in the vacuum
chamber and the sample are set at temperature 26.5°C. When the
pump turns on, the temperature measured by the thermistor at
the metal plate rises quickly to 29.5°C (P = 1.6 W, @ 1940 nm).
Keeping the pump power unchanged, the temperature versus
time was measured and recorded from 17:30 to 22:30, as shown
in Fig. 3.16. Closer inspection of the results in Fig. 3.16 indicates a
temperature difference between the sample and the inside of the
vacuum chamber that is consistent with cooling. The temperature in
the sample drops more than 0.6°C. We note that the input power is
1.6 W and the pump residue is about 0.25 W, meaning that the fiber
absorbs (including loss) 1.35 W of pump power continuously during
that time. We have carried out cooling experiments using the second
cooling fiber Tm-CF2 and the same configuration as in Fig. 3.15. The
temperature evolution for the fiber Tm-CF2 is similar to Fig. 3.16,
but the magnitude of temperature in the sample drops more than
1.2°C versus 0.6°C in first fiber Tm-CF1.
Sample Temp. (°C)
Chamber Temp. (°C)
Figure 3.16 Temperature versus time measured in vacuum chamber with

pump power 1.6 W at 1940 nm. The curves show temperatures
measured at the sample attached to the cooling fiber and inside
the vacuum chamber. Reprinted with permission from Ref.
Engineers.
Qualitatively, the above results of experiments using thermistors

to measure temperature in the fiber indicate cooling effects could
have occurred in the fiber. It is important to point out here that
there are two thermistors in the experiments, and they consume
power and therefore generate and contribute heat to the sample.
To determine precisely the temperature in the fiber under laser
pumping, a noncontact measurement approach has been developed
and is presented next.
3.3.3.2 Noncontact measurements

As discussed above, the direct contact measurement approach uses
thermistors. The results of direct contact measurements indicate
qualitatively that optical cooling could occur in our cooling fiber
under laser pumping. To confirm the effect quantitatively, we
have developed experimentally and theoretically a noncontact
measurement approach. The noncontact method investigates the
fluorescence spectrum of an isolated fiber under two conditions: (i)
The fluorescence spectra of the Tm-doped glass fiber are measured
under different pump powers (P-dependent spectra), and (ii) the
fluorescence spectra of the Tm-doped glass fiber are measured
under low pump power and different temperatures (T-dependent
spectra), which are set by a thermoelectric cooler (TEC) attached to
the fiber. Figure 3.17 shows configurations of the experiments for
measuring P-dependent and T-dependent spectra [10, 11].
Note that the configurations shown in Fig. 3.17 are quite simple
in the fiber system using beam combiners. Figure 3.18 shows the
power-dependent and temperature-dependent fluorescence spectra
under 1950 nm pumping.
We note that the results show similarities between the
P-dependent and T-dependent fluorescence spectra from 1600 nm
to 1920 nm, which is shorter than the pump wavelength 1950 nm
(i.e., the required condition for the optical cooling process). This is
consistent with the interpretation that the Tm-doped fiber under
different pumping powers undergoes optical cooling processes,
and the emitted fluorescence spectra with higher pump power
correspond to lower temperature spectra in T-dependent ones. For
example, the lowest pump power spectrum (~0.9 W) corresponds
to the one at 20°C. Significantly, the results in Fig. 3.19 show that the
temperature in the fiber under pump power 2.55 W corresponds to

–10°C, and the temperature under power 2.9 W is even lower. Note
that, in the present experiment, we only a TEC that can drive the
temperature to only –10°C.
fluorescence
fluorescence
Pump laser 3cm 3cm

Pump
1940nmlaser
laser
105/125
Beam combiner NA=0.2 Beam combiner
TEC
Figure 3.17 Configurations for spectroscopic measurements. Left:

backward measurement of pump power-dependent spectra.
Right: backward measurement of temperature-dependent
spectra of the Tm-doped cooling fiber (blue section).
1.0 0.90 W 1.0 -10 C

Normalized Intensity
1.45 W -5C
0.9 2.05 W 0.9 0C

0.8 2.55 W 0.8 +5C
2.90 W +15 C
0.7 0.7 +20 C
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 P-dependent 0.2 T-dependent
0.1 spectrum 0.1 spectrum
0.0 0.0
1600 1640 1680 1720 1760 1800 1840 1880 1920 1600 1640 1680 1720 1760 1800 1840 1880 1920
Wavelength (nm) Wavelength (nm)
Figure 3.18 Power (P)-dependent and temperature (T)-dependent spectra

under 1950 nm pumping.
1.0 T = -10 C
0.9 P = 2.55 W -10 C
P = 2.90 W
0.8
0.7
0.6 From room temperature
0.5
Temperature drops
0.4
to - 10C (P=2.55 W)
0.3
0.2
0.1 P=2.90 W
0.0 T < -10 C
1600 1640 1680 1720 1760 1800 1840 1880 1920
Wavelength (nm)
Figure 3.19 Comparison of P-dependent and T-dependent spectra under

1950 nm pumping. Temperature in the fiber under pumping
with P = 2.55 W drops from room temperature ~24°C to about
–10°C. Temperature under pumping with P = 2.90 W is even
lower than –10°C.
We also have performed theoretical modeling of the Tm-doped

glass and the theoretical results are in good agreement with the
experiments (see Fig. 3.20). However, the theoretical calculations
are preliminary as several material parameters in the tellurite
glass are not well determined. Furthermore, real conditions in
the experiments are not fully taken into account in the theoretical
modeling. Those issues need to be addressed in the future.
experiment theory
1.0 1.0
0.90 W
0.9 1.45 W
2.05 W
0.8 2.55 W 0.8

2.90 W
0.7
0.6 0.6
0.5 285 K
0.4
285
290 KK
0.4 290
295 KK
0.3 295
300 KK
0.2 300
305 KK
0.2 305
0.1 310 KK
310
315 KK
0.0
1600 1640 1680 1720 1760 1800 1840 1880 1920 0.0
Wavelength (nm) 1600 1640 1680 1720 1760 1800 1840 1880 1920
Wavelength [nm]
Figure 3.20 We find good agreement between experimental (left)

and theoretical spectra (right), comparing power- and
temperature-dependent spectra. The results are consistent
with the interpretation that the temperature in the fiber drops
more than 30°C from room temperature.
1.0 +24 C
1.0 0.18 W 0.9 +12 C
0.9 0.45 W +5C
0.65 W 0.8 0C
0.8 0.90 W -5C
1.20 W 0.7 -10 C
0.7
0.6
0.6
0.5
0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
1600 1640 1680 1720 1760 1800 1840 1880 1920 1600 1640 1680 1720 1760 1800 1840 1880 1920
Wavelength (nm) Wavelength (nm)
Figure 3.21 P-dependent and T-dependent spectra under 1940 nm

pumping.
Figure 3.21 shows the results for the 1940 nm pump laser. Again,
the results are consistent with the interpretation that net optical
cooling in the fiber has been achieved. Preliminary analysis shows

that the temperature drops about 30°C from room temperature
with 1940 nm pump power P = 1.2 W. We believe that this is the
first experimental indication of possible optical net cooling in the
Tm-doped tellurite glass fiber. In the past, cooling in the Yb-doped
ZBLAN fiber has been demonstrated [10, 11]. Note that the ZBLAN
glass fiber is fracture-prone compared to tellurite glass. As a result,
it is difficult to have an all-fiber integrated system for optical cooling
based on the ZBLAN glass fiber. Optical cooling has also been
demonstrated experimentally in Tm-doped ZBLAN glass [9] but not
in the fiber.
Acknowledgments
Elements of this work were sponsored by the Office of Naval Research
(ONR) under contract/grant STTR N00014-11-C-0477, ONR Public
Release Distribution Statement A, “Approved for public release;
distribution is unlimited.” Additional support was provided by the
Air Force Office of Scientific Research (AFOSR) under contract/grant
FA9550-11-0048 and by TRIF Photonics. We thank G. Hardesty for
his contributions to our theoretical effort.
References

Lumineszenz- und Temperaturstrahlung, Z. Phys., p. 739.
2. Epstein, R. I., Buchwald, M. L., Edwards, B. C., Gosnell, T. R., Mungan,
C. E. (1995). Observation of laser-induced fluorescent cooling of a
solid, Nature, 377, pp. 500–503.
5. Sheik-Bahae, M., Epstein, R. I. (2007). Optical refrigeration, Nat.
Photonics, 1, pp. 693–699.
Phys., 73, p. 086501.
References 109
7. Nguyen, D. T., Shanor, C., Zong, J., Tian, W., Yao, Z., Wu, J., Weiss, J.,
Binder, R., Chavez-Pirson, A. (2011). Conceptual study of a fiber-optical
approach to solid-state laser cooling, Proc. SPIE, 7951, p. 795109.
8. Epstein, R. I., Sheik-Bahae, M., eds. (2009). Optical Refrigeration
(WILEY-VCH Verlag, Germany).
9. Hoyt, C. W., Hasselbeck, M., Sheik-Bahae, M., Epstein, R. I., Greenfield,
T. R. (1997). Laser cooling of a solid by 16 K starting from room
temperature, Phys. Rev. Lett., 78, pp. 1030–1033.
11. Gosnell, T. R. (1998). Laser cooling of a solid by 16 K starting from
room temperature, Opt. Lett., 24, pp. 1041–1043.
12. Nguyen, D. T., Thapa, R., Rhonehouse, D., Zong, J., Miller, A., Kwong,
N. H., Binder, R., Chavez-Pirson, A. (2013). Towards all-fiber optical
coolers using Tm-doped glass fibers, Proc. SPIE, 8638, p. 86380G.
13. McCumber, D. E. (1964). Einstein relations connecting broadband
emission and absorption spectra, Phys. Rev., 136, pp. A954–A957.
14. van Roosbroeck, W., Shockley, W. (1954). Photon-radiative
recombination of electrons and holes in germanium, Phys. Rev., 94, pp.
1558–1560.
15. Zimmermann, R. (1987). Many-Particle Theory of Highly Excited
Semiconductors (Teubner, Leipzig).
16. Kwong, N. H., Rupper, G., Binder, R. (2009). Self-consistent T-matrix
theory of light-absorption and luminescence, Phys. Rev. B, 79, p.
155205.
17. Richtmyer, R. D., Morton, K. W. (1967). Difference Methods for Initial-
Value Problems, 2nd Ed. (Interscience Publishers, New York).
18. Nguyen, D. T., Zong, J., Rhonenhouse, D., Yao, Z., Hardesty, G., Kwong,
N. H., Binder, R., Chavez-Pirson, A. (2012). All fiber approach to solid-
state laser cooling, Proc. SPIE, 8275, p. 827506.
19. Hehlen, M. P., Epstein, R. I., Inoue, H. (2006). Laser cooling in rare-earth
doped solids: the next step, Proc. SPIE, 6130, p. 61300N.
Chapter 4
Laser Cooling of Solids around 2.07

Microns: A Theoretical Investigation
Guang-Zong Dong and Xin-Lu Zhang

Key Laboratory of In-Fiber Integrated Optics (Ministry of Education) and
College of Science, Harbin Engineering University, Harbin, 150001, P. R. China
zhangxinlu1@aliyun.com
4.1 Introduction
Since the first demonstration provided by Epstein et al. [1] two
decades ago, laser cooling of solids or optical refrigeration has
been developed systematically [2–4], and recent advances have
claimed cryogenic cooling in highly purified Yb3+-doped fluoride
crystals [5–8]. The simple energy level structure together with the
wide energy gap imparts Yb3+ ion–doped sample an extremely high
quantum efficiency, making Yb3+ ion an ideal choice for cryogenic
optical refrigeration. Nevertheless, other trivalent rare earth ions,
such as Tm3+, Er3+, and Ho3+, attract increasing interest due to two
main reasons. For one thing, the operation wavelength region for
optical refrigeration can be expanded by adopting these ions, which
could be technically important in practical applications. For another,

www.panstanford.com
112 Laser Cooling of Solids around 2.07 Microns
it is possible to introduce new cooling mechanics to multimanifold

cooling systems to improve the cooling performance [9–12]. Since
cooling around 1.5 and 2.0 μm has been achieved in Er3+- and Tm3+-
doped solids [13–16], it is wondered whether optical refrigeration
can be extended beyond 2 μm in Ho3+-doped solids.
In 2013, Bowman et al. experimentally investigated the cooling
performances of various crystals and glasses doped with Ho3+ ion
[17]. Results show that only a 0.3% Ho3+:ZBLANP glass sample tends
to exhibit refrigeration when pump wavelength exceeds 2.07 μm. The
poor cooling performance of Ho3+-doped solids is attributed to three
aspects: multiphonons, background absorption, and fluorescence
quenching induced by energy transfer upconversion (ETU). To
point out, the multiphonon and background absorption issues can
be solved (or partly solved) by adopting highly purified, low-host-
phonon materials. Specially, the corresponding energy gap should
exceed eight times the maximum host phonon energy to ensure
acceptable high quantum efficiency for cooling [18]. The influence
of upconversion on optical refrigeration was initially mentioned by
Hoyt et al. when Tm3+-doped glass was first used to demonstrate net
cooling [13]. The heating or cooling effect caused by upconversion
is estimated negligible, and the conventional two-manifold cooling
model fits the cooling result well. On the other hand, Garcia-Adeva et
al. suggested that excited-state absorption (ESA) and ETU processes
in Er3+-doped KPb2X5 (X = Cl, Br) crystals could be utilized as
additional cooling channels since the nonradiative transitions are
largely averted [9].
For optical refrigeration with Ho3+-doped solids, the ETU process
is typically strong, and the transitions from the third and second
excited states to the lower nearby states are mainly nonradiative,
so the heat generated by the fluorescence quenching effect is not
negligible [19]. To describe laser cooling of Ho3+-doped solids
properly, a cooling model with consideration of upconversion will be
presented in Section 4.2. Simulation based on the generalized cooling
model will reveal the requirements for achieving net cooling in
Ho3+-doped fluoride crystals and the potential cooling performance.
Moreover, the role of upconversion in optical refrigeration with
multimanifold rare earth ions is comprehensively investigated to
obtain a common criterion on whether cooling can be improved
through an upconversion channel.
Conventional Laser Cooling of Holmium-Doped Fluoride Crystals 113
To overcome the ETU-induced heating and realize efficient

and enhanced laser cooling in Ho3+-doped solids, in Section 4.3,
we will provide a modified excitation scheme based on ground-
state absorption (GSA) of 5I8 Æ 5I7 at around 2.07 μm and ESA of
5I Æ 5F at around 2.28 μm. Compared with the conventional
5 5
pumping scheme, the co-pumping scheme ameliorates the cooling
performance significantly through alleviating the upconversion-
quenching-induced heat [12], that is, the heat-producing 5I5 Ho3+
ions are pumped to the 5F5 state by the laser at 2.28 μm. Fortunately,
the 5F5 Ho3+ ions barely transit to the lower nearby state and
tend to decay radiatively. Therefore, major heat produced in the
conventional laser cooling of Ho3+-doped solids is avoided, and both
cooling efficiency and cooling power density can be improved.
In Section 4.4, we broaden the research domain of optical
refrigeration to rare earth ion co-doped systems. Specifically, when
Ho3+–Tm3+ co-doped YLiF4 crystals are pumped at around 2.07 μm,
Ho3+ ions will populate at the 5I7 manifold. Meanwhile, Tm3+ ions
will also be sensitized by absorbing energy transferred from 5I7 Ho3+
ions. Since the average energy of fluorescence photons generated by
Tm3+ 3F4 Æ 3H6 radiative transitions is higher than that generated
by Ho3+ 5I7 Æ 5I8 radiative transitions, energy transfer (ET) from
Ho3+ to Tm3+ ions will induce annihilation of host phonons, and
bring enhancement in both the cooling efficiency and the cooling
power. Therefore, this cooling mechanism is referred next as energy
transfer–enhanced laser cooling (ETLC) of solids [11].
4.2 Conventional Laser Cooling of Holmium-

Doped Fluoride Crystals
The main upconversion process and the anti-Stokes fluorescence
cooling (ASFC) processes of Ho3+-doped fluoride crystals are
illustrated in Fig. 4.1. The ground-state (5I8) Ho3+ ions are pumped to
the first excited state (5I7) with laser excitations of energy hν (solid
red arrow on the left) and then relax thermally among the Stark
levels in a timescale of picoseconds with host phonon absorption.
Part of the Ho3+ 5I7 ions de-excite by emitting fluorescence photons
of mean energy hν10 (solid blue arrow, ν10 > ν), which accounts for
the ASFC process. The rest of the 5I7 Ho3+ ions de-excited and transfer
their energy to the neighboring 5I7 Ho3+ ions, causing population
accumulation at the third excited state (5I5) with host phonon
absorption (solid green arrows). Comparison between the energy
mismatch and the host phonon energy (for YLiF4 crystal host, ħω ≈
450 cm–1) indicates that about two host phonons are annihilated for
each ETU process (see Table 4.1).
Figure 4.1 Schematic of optical refrigeration in Ho3+-doped fluoride

crystal with consideration of ETU. The level labels used in Eq.
4.1 are also indicated.
The ET processes intrinsically appear in pairs, that is, the

reverse ET process, 5I5 Æ 5I7; 5I8 Æ 5I7 causes Ho3+ ions populating
at the 5I7 state. However, this ET process is largely averted since the
5I Ho3+ ions relax fast (nonradiatively) to the 5I state in fluoride
5 6
crystals [22]. Other ET processes, such as 5I7 Æ 5I8; 5I7 Æ 5I6 and 5I6
Æ 5I7; 5I7 Æ 5I8, are negligible due to the large energy mismatches
[21, 22]. It should be noted that 5I7 Ho3+ ions can also transit to the
5I state by absorbing laser excitations and surrounding phonons
5
(phonon-assisted ESA), whereas the pump-wavelength-dependent
ESA process is insignificant compared to the 5I7 Æ 5I5; 5I7 Æ 5I8
ETU process, since more than three phonons are needed in such a
transition process when the pump wavelength is tuned to the long-
wavelength tail of the 5I7 absorption spectrum (see Fig. 4.2). The 5I5
and 5I6 Ho3+ ions de-excite to the lower manifolds either radiatively
(solid purple and brown arrows), eliminating heat from the host
(upconversion-assisted cooling), or nonradiatively (dashed gray
arrows on the right), introducing extra heat to the host.
Table 4.1 Branching ratios, lifetimes, and energy gaps in Ho3+:YLiF4
Transitions βij τr (ms) εij (cm–1)

5I
7 Æ 5I8 1 13.68 4838
5I
6 Æ 5I
8 0.9098 6.01
5I
6 Æ 5I
7 0.0902 3377
5I
5 Æ 5I
8 0.4002 7.37
5I
5 Æ 5I
7 0.5649
5I
5 Æ 5I
6 0.0348 2444
5F
5 Æ 5I
8 0.7738 0.554
5F
5 Æ 5I
7 0.1813
5F
5 Æ 5I
6 0.0417
5F
5 Æ 5I
5 0.0032 4152
5F
5 Æ 5I
4 0.0000
Source: Refs. 20 and 21.
With consideration of the main ETU process, the population

evaluation of each manifold can be described by using the following
rate equations:
dn3
= -( b30W30 + b31W31 + b32W32 )n3 + p11 (C )n1 (4.1a)
2
dt
dn2
= -( b20W20 + b21W21 )n2 + b32W32 n3 (4.1b)
dt
dn1 I
= s abs (n )(n0 - n1 ) + b31W31 n3 + b21W21 n2
dt hn (4.1c)
-W10 n1 - 2p11 (C )n
2
1
Â n = CN (4.1d)
i
i =0
where ni (i = 0, 1, 2, 3) is the population density of the ith level

(see Fig. 4.1). I is the intensity of the laser beam at pump frequency
ν. σabs is the absorption cross section of the Ho3+ 5I7 manifold. C is
the Ho3+ dopant concentration and N is the total number density of

substitutable sites of the host (for a YLiF4 crystal, N = 1.396 × 1022
ions/cm3 [23]). βij (i > j, i = 1, 2, 3, j = 0, 1, 2) is the branch ratio for the
i Æ j transition. The total effective decay rate is expressed as
Wij = ηe,ij Wij,r + Wij,nr (4.2)
where Wij,r and Wij,nr are the radiative and nonradiative decay rates,
respectively. ηe,ij is the corresponding fraction of photons escaping
from the cooling sample [24]. p11(C) is the dopant-concentration-
dependent ETU parameter. Considering a dipole–dipole interaction
in the frame of Dexter’s ET theory, the microscopic ET rate depends
on the separation of the interacting ions to the inverse sixth power
[25]. Hence, the ETU parameter is approximately proportional to
the second order of the ion concentration. The ETU parameter is
also temperature dependent, since host phonons participate in the
upconversion process [26]. Therefore,
p11(C) µ C2n(T)2 (4.3)
where n = [exp(ℏωmax/kBT) – 1]–1 is the Bose–Einstein occupation
number, with ℏωmax being the maximum host phonon energy. Take
the 2% Ho3+: YLiF4 crystal, for example; the ETU parameter at room
temperature is found to be 3.2 × 10–24 m3∙s–1 [22].
The total absorption includes the resonant absorption and the
background absorption
Pads = Pr + Pb = σabs(v)(n0s – n1s)I + αbI (4.4)
where αb is the background absorption coefficient that can be treated
as a constant [24]. The power density of the fluorescence generated
by transitions from the first, second, and third excited states to the
lower energy state(s) can be described, respectively, as
P1 = ηe,10W10,rhv10n1s (4.5a)
P2 = ηe,20β 20W20,rhv20n2s (4.5b)
P3 = ηe,30β30W30,rhv30n3s + ηe,31β31W31,rhv31n3s (4.5c)

where nis is the steady-state population of ith energy level that can
be calculated from Eq. 4.1 by setting dni/dt = 0. νij is the average
fluorescence frequency corresponding to the i Æ j transition (as
shown in Fig. 4.1). The cooling power density can be expressed by
using the energy conservation law
Pcool = P1 + P2 + P3 – Pads (4.6)

The cooling efficiency is defined as
ηcool = Pcool/Pads (4.7)
Analytical expressions of the cooling power density and the
cooling efficiency in compact form are hard to present, as quadratic
terms emerge in Eq. 4.1, whereas numerical results are easily
obtained once detailed spectrum parameters are available.
It is important to mention that a two-manifold treatment of
cooling with a Ho3+-doped solid is still valid, given that the second
and third excited states recombine predominantly nonradiatively.
Neglecting the radiative terms on the right-hand side of Eqs. 4.1a
and 4.1b, the population evaluation of the 5I7 manifold can be
approximately expressed as
dn1 I
= s abs (n )(n0 - n1 ) - W10 n1 - p11 (C )n1
2
(4.8)
dt hn
Comparing with the traditional two-manifold cooling model
that is widely used for describing optical refrigeration with Yb3+-,
Tm3+-, and Er3+-doped solids [2], it can be seen that only a dopant-
concentration-dependent upconversion term is supplemented in the
above equation. The heat generated from ETU can be approximately
expressed as
p11 (C ) n1s hn = b21W21,nr e 21 n2s + b32W32,nr e 32 n3s (4.9)
2
where εij is the energy gap between i and j levels. By using Eqs.
4.4–4.7, the approximate expression of the cooling efficiency can be
written as
n 10
¢ = hext
hcool ¢ habs - 1 (4.10)
n
where
-1
habs = [1 + a b s abs (n0 s - n1 s )] (4.11)
is the pump intensity– and frequency-dependent absorption
efficiency [5]. In Eq. 4.10, we have introduced the equivalent external
quantum efficiency
-1
¢ = (1 + W10¢ ,nr W10 ,r )
hext (4.12)
where
¢ = W10,nr + p11 (C )n1 s (4.13)
W10,nr
is the equivalent nonradiative transition rate that characterizes

the quantum defect caused by both nonradiative transition from
the first excited state and ETU-induced fluorescence quenching.
It can be seen from Eq. 4.3 that the equivalent external quantum
efficiency decreases with increasing the dopant concentration. On
the contrary, the absorption efficiency increases with increasing
the dopant concentration. To achieve net cooling (η¢cool > 0), the
condition
n
¢ (C )habs (C ) >
hext (4.14)
n 10
needs to be satisfied. By using Eq. 4.14, the appropriate dopant
concentration for a Ho3+ cooling system should be roughly figured
out. So the two-manifold model could still provide approximate
prediction on the cooling performance of Ho3+-doped solids, in
which the transitions from the higher manifolds are predominantly
nonradiatively, and given that the external quantum efficiency is
substituted by Eq. 4.12.
The cooling system benefits from the ETU channel only when
the overall number of phonons annihilated exceeds that created by
the following transitions from the third and second excited states
to the lower energy states. Otherwise, the ETU channel harms the
cooling performance. For the 5I7 Æ 5I8; 5I7 Æ 5I5 ET process, the 5I7
population is reduced by two, and two laser photons (Eu-a = 2hν)
originally served as excitations of the conventional cooling channel
are absorbed (with host phonon annihilation); as a result, the 5I5
population is increased by one. The 5I5 Ho3+ ions de-excite to the
ground state through various transitions. If we define the weighting
parameters that characterize the probability of each radiative and
nonradiative transitions as
-1
Ê i -1
ˆ
hij = he , ij b ij Wij ,r Á
Ë
Â b ik Wik ˜ (4.15a)
¯
k =0
-1
Ê i -1
ˆ
xij = b ij Wij ,nr Á
Ë
Â b ik Wik ˜ (4.15b)
¯
k =0
where i = 1, 2, 3, j = 0, 1, 2, and i > j. The fluorescence energy and

the heat generated after absorbing the two laser photons can be
described, respectively, as
E u-f = h30 hn 30 + h31 hn 31 + x32h20 hn 20 + (h31 + x32x21 )h10 hn 10 (4.16a)
E ETU = x32 e 32 + x32x21 e 21 + (h31 + x32x21 )x10 e 10

(4.16b)
ª x32 e 32 + x32x21 e 21
To obtain the last equality on the right-hand side of Eq. 4.16b,

the relation ξ10 << ξ21(32) (the nonradiative transition rate of the
5I Æ 5I transition is far smaller than that of the 5I (5I ) Æ 5I (5I )
7 8 5 6 6 7
transition) has been used. The criterion for cooling with the ETU
channel is expressed as
Eu–f – Eu–a(2hv) > EETU (4.17)
Substituting Eq. 4.16 into Eq. 4.17, the pump frequency range for
cooling with the ETU channel is obtained,
v < vu (4.18)
where the cutoff frequency is written as
n u = h30n 30 + h31n 31 + x32h20n 20 + (h31h10 + x32x21h10 )n 10
(4.19)
-x32 f32 - x32x21 f21 ,
and fij = εij/h is the equivalent frequency of the corresponding

energy gap. Since the absorbed photon energy for each ESA process
is 2hν, and the transition rates are almost independent of the pump
condition, Eqs. 4.18 and 4.19 also indicate the pump frequency
range for cooling through the ESA channel. If the cutoff frequency
locals outside the absorption spectrum range of the 5I7 manifold,
upconversion cannot assist the cooling process. For cooling samples
strongly affected by upconversion quenching, the corresponding
nonradiative processes (minus terms on the right-hand side of Eq.
4.19) are stronger than the radiative processes (plus terms), so the
value of the cutoff frequency is essentially negative. In such a case, no
matter how the pump frequency is red-detuned, the upconversion
processes barely obstruct the cooling process. It should be noted
that, rare earth ions of a multimanifold exhibit similar transition
properties when pumped to their first excited state. Therefore, the
criterion is widely applicable in optical refrigeration with rare earth
ions of a multimanifold.
To simulate the cooling performance of conventional Ho3+ optical

refrigeration, we need the spectrum parameters of the Ho3+:YLiF4
crystal, summarized in Table 4.1. The nonradiative transition rates
can be obtained by using the energy gap law [27],
Wnr,ij = β(1 + n)m exp(–αεij) (4.20)
where α and β are the host material–dependent characteristic
constants. For a YLiF4 crystal, α = 3.8 × 10–3 cm and β = 3.5 × 107 s–1
[28]. m = εij/ℏωmax is the number of phonons needed to bridge the
energy gap of εij. At 300 K, the nonradiative transition rates of 5I7 Æ
5I , 5I Æ 5I , and 5I Æ 5I transitions are calculated as 1.36, 234.77,
8 6 7 5 6
and 6308.91 s–1, respectively. Other nonradiative transition rates
are comparatively small and are assumed to be zero in the following
simulations. The emission cross sections of 5I7, 5I6, and 5I5 manifolds
are shown in Fig. 4.2, and the average wavelengths of fluorescence
generated by 5I7 Æ 5I8, 5I6 Æ 5I8, and 5I5 Æ 5I8 transitions are found
to be 2014, 1185, and 908 nm, respectively. The average wavelength
of fluorescence generated by the 5I5 Æ 5I7 transition is estimated to
be 1675 nm [21, 29].
Figure 4.2 (a) Absorption and emission cross section (π-pol.) of Ho3+
5I manifold; emission cross section (π-pol.) of Ho3+ 5I (b)
7 6
and 5I5 (c) manifolds; and (d) absorption and emission cross
section (π-pol.) of Tm3+ 5F4 manifold in YLiF4 host matrix
at room temperature [20, 23] (the Tm3+ spectrum will be
needed in Section 4.4). The dashed lines indicate the average
fluorescence wavelengths.
Suppose the cooling sample is geometrically optimized, so

fluorescence photons totally escape from the cooling sample (ηe,ij =
1). Calculating the result of Eq. 4.19 shows that for the Ho3+:YLiF4
crystal, the cutoff frequency is negative; therefore, the ETU process
of the Ho3+:YLiF4 crystal intrinsically harms its cooling performance.
As Ref. [17] indicates, to realize net cooling of Ho3+-doped solids,
it is essential to inhibit ETU. For this reason, diluting the dopant
could be a feasible solution. To figure out the appropriate dopant
concentration region for optical refrigeration with the Ho3+:YLiF4
crystal, the cooling power density and the cooling efficiency that
depend on the dopant concentration are simulated below.
From Fig. 4.2a we know that it is rational to select 2070 nm as
the pump wavelength. The cooling power density (Eq. 4.6) and the
cooling efficiency (Eq. 4.7) as functions of pump intensity and dopant
concentration are depicted in Fig. 4.3 (the background absorption
is assumed to be 4.0 × 10–4 cm–1, which is currently practical for
manufacturing [6]). It can be seen that net cooling would be arduous
to achieve if the selected YLiF4 crystal sample is doped with more
than 1% Ho3+ ions, since the cooling region for pump intensity is
considerably narrow and the achievable cooling power density
is quite low at such a high concentration level. In view of heating
induced by the upconversion quenching, it is rational to limit the
dopant concentration within 1%. On the other hand, the radiation
center (Ho3+ ion) is desired to be densely doped to provide enough
cooling, surpassing the background absorption heating. A more
practical dopant concentration range for the Ho3+:YLiF4 crystal with
a background absorption of 4.0 × 10–4 cm–1 is suggested as 0.2%–
1%. For each pump wavelength, there exists an optimal dopant
concentration from the perspective of maximizing the cooling power
density. We figure out that at a pump wavelength of 2070 nm, the
optimal dopant concentration for the Ho3+:YLiF4 crystal is 0.54%,
the maximum achievable cooling power density is 9.7 × 104 W∙m–3,
and the corresponding cooling efficiency is 0.29%.
Figure 4.4 shows the comparison between the cooling results
predicted by the conventional two-manifold cooling model without
consideration of ETU (M1), the generalized cooling model (M2), and
the two-manifold cooling model with consideration of ETU (M3). At
low pump intensity (I < 106 W∙m–2 for a 0.54% Ho3+:YLiF4 crystal),
the 5I7 state is sparsely populated, so the ETU process is weak, and all
of these models are fit to describe the cooling performance of Ho3+-

doped fluoride crystals. However, as the pump intensity increases,
M1 tends to overestimate the cooling results, because the heating
process induced by upconversion quenching is omitted. On the
contrary, M3 tends to underestimate the cooling results because the
phonon annihilation induced by phonon-assisted ETU is excluded.
Figure 4.3 (a) The cooling power density and (b) the cooling efficiency of
a Ho3+:YLiF4 crystal as functions of pump intensity and dopant
concentration. The pump wavelength is 2070 nm, and the
background absorption is 4.0 × 10–4 cm–1.
Figure 4.4 Comparison between M1 (dotted curves), M2 (solid curves),

and M3 (dashed curves). (a) The cooling power density and (b)
the cooling efficiency. The dopant concentration is 0.54%, the
pump wavelength is 2070 nm, and the background absorption
is 4.0 × 10–4 cm–1.
Figure 4.5 shows the maximum attainable cooling power density

and the corresponding optimal dopant concentration as functions of
pump wavelength. It can be seen from Fig. 4.5a that for a YLiF4 crystal
Efficient and Enhanced Holmium Optical Refrigeration via Co-Pumping 123
with a background absorption of 4.0 × 10–4 cm–1, the ideal pump

wavelength is around 2067 nm. A cooling power density of about 1.1
× 105 W∙m–3 is expected in ~0.47% Ho3+-doped samples. Purifying
the host is an important way to improve the cooling performance. As
it can be seen in Fig. 4.5b, if the background absorption is reduced by
an order of magnitude, the maximum cooling power density (MCPD)
could exceed 7.5 × 105 W∙m–3 in ~0.41% Ho3+-doped samples, and
the corresponding optimal pump wavelength red-shifts to ~2083
nm. In fact, the optimal dopant concentration is determined by
a balancing consideration of heat loads induced by background
absorption and fluorescence quenching. As the background
absorption decreases, the parasitic heat problem is alleviated and
the weight of heating induced by fluorescence quenching effect
increases. A reduced dopant concentration essentially weakens
the ETU process. Therefore, the optimal dopant concentration for a
host with lower background absorption is lower than that for a host
with higher background absorption. For optical refrigeration with a
highly purified sample, it is possible to tune the pump wavelength to
the red-tail part of the absorption spectrum (where the cross section
is tiny) to obtain a high cooling power density.
MCPD
MCPD
ODC ODC
Figure 4.5 The maximum cooling power density (MCPD) and the corre-
sponding optimal dopant concentration (ODC) verses pump
wavelength at background absorptions of (a) 4.0 × 10–4 cm–1
and (b) 4.0 × 10–5 cm–1.
4.3 Efficient and Enhanced Holmium Optical

Refrigeration via Co-Pumping
Upconversion quenching makes the attempt of cooling Ho3+-doped
fluoride crystals a challenging task. When pumped at around 2.07
μm for realizing conventional ASFC, Ho3+ ions will populate at the

5I manifold due to an ETU of 5I Æ 5I ; 5I Æ 5I . The following
5 7 8 7 5
transition of 5I5 Æ 5I6 is predominantly nonradiative, preventing
effective cooling. To solve the upconversion-induced heating issue,
it is possible to adopt the co-pumping scheme illustrated in Fig. 4.6.
First of all, Ho3+ ions are pumped from the top of the ground-state
manifold (5I8) to the bottom of the first excited-state manifold (5I7)
by laser at a frequency of νg to facilitate the ASFC process (solid red
and green arrows below). Part of 5I7 Ho3+ ions interact through an
electromagnetic multipole, resulting in an ETU of 5I7 Æ 5I8; 5I7 Æ
5I (solid yellow arrows) with two host phonons’ absorption (solid
5
gray arrow). Among all the transitions from 5I5 and 5I6 to the lower
manifolds, 5I5 Æ 5I6 and 5I6 Æ 5I7 are predominantly nonradiative
due to the small energy gaps, which bring main heat to the system
(see Section 4.2).
To cut down the heat induced by upconversion quenching, the
laser at a frequency of νe that pumps 5I5 Ho3+ ions to the 5F5 manifold
(solid brown arrow above) may be added to the system. The branch
ratio of transition from 5F5 to its lower adjacent manifold (5I4) is
approximately zero (see Table 4.1). The branch ratio for the 5F5 Æ 5I5
transition is small, and the corresponding energy gap is much wider
than 5I5–5I6. According to the energy gap law, the 5F5 manifold should
be less heat involving than the 5I5 manifold. So, the ESA of 5I5 Æ 5F5
that aims at depopulating the 5I5 manifold may improve the cooling
performance. To confirm this guess, a quantitative comparison of
heat production through different transition channels is needed.
In assessing the heating process of a particular transition
channel, it would be convenient to utilize the weighting parameters
(see Section 4.2) that characterize the probability of each radiative
and nonradiative transition. The heat that generated after each 5I7 Æ
5I ; 5I Æ 5I ETU process can be described by Eq. 4.16b. Similarly,
8 7 5
the heat that generated after each 5I5 Æ 5F5 ESA transition can be
expressed as
E ESA ª h43x32 e 32 + (h42 + h43x32 )x21 e 21 (4.21)
If EESA is smaller than EETU in value, we can draw the conclusion

that the 5I5 Æ 5F5 ESA process dwindles ETU-induced heat. Indeed,
the calculation based on Eqs. 4.16b, 4.20, and 4.21 shows that EESA is
only about 0.3% of EETU at 300 K. Therefore, the co-pumping scheme

could improve the cooling performance of Ho3+-doped solids.
As shown in Fig. 4.6, the rate equations describing the population
dynamics of the cooling system are given by
dn4 Ie
= s e (n e )(n3 - n4 ) - A4 (4.22a)
dt hn e
dn3 Ie
= -s e (n e )(n3 - n4 ) - A3 + B3 + p11 (C )n1 (4.22b)
2
dt hn e
dn2
= - A2 + B2 (4.22c)
dt
dn1 Ig
= s g (n g )(n0 - n1 ) - A1 + B1 - 2p11 (C )n1 (4.22d)
2
dt hn g
Â n = CN (4.22e)
i
i =0
where ni (i = 0, …, 4) is the population density of the ith level (see

Fig. 4.6). Ig and Ie are the intensities of the incident laser beams at
the frequency νg and νe, respectively. σg and σe are the temperature-
dependent GSA cross section of the 5I8 Æ 5I7 transition and the ESA
cross section of the 5I5 Æ 5F5 transition, respectively, and according
to the reciprocal relation [30], their temperature dependence can be
determined by
u
Z g(e) Ê hn g(e) - E g(e)
ZL
ˆ
s g(e) (T ) µ exp Á ˜ (4.23)
Ë k BT ¯
l
Z g(e)
where Zug(e) and Z g(e)

l are partition functions of 5F5(5I7) and 5I5(5I8)
manifolds. EZL 5 5
g(e) is the zero-line energy of the F5( I7) manifold.
i -1
Ai = Â b W n (4.24)ij ij i
j =0
is the sum of all decay terms from the ith level (composed of radiative
and nonradiative transitions), and
4
Bi = Â b W n (4.25) ji ji j
j = i +1
is the sum of all gain terms to the ith level.

Figure 4.6 Schematic of the co-pumping scheme for efficient optical

refrigeration in Ho3+-doped fluoride crystals. The transition
of 5F5 Æ 5I4 is omitted due to the tiny branch ratio. The level
labels used in Eq. 4.22 are also indicated.
The total absorption includes the resonant absorptions (GSA and

ESA) and the background absorption,
Pabs = σg(n0s – n1s)Ig + σe(n3s – n4s)Ie + αb(Ig + Ie) (4.26)
where nis is the steady-state population of the ith level. The total
emitted fluorescence power density can be described as
4 i -1
Pf = Â Âh e , ij
b ij Wij nis hn ij (4.27)
i =1 j =0
where νij is the mean fluorescence frequency for the i Æ j transition.

The cooling power density is obtained from the energy conservation
law
Pcool = Pf – Pads (4.28)
It is essential to note that while the upconversion quenching–
induced heat is alleviated with the ESA transition, the background
absorption is increased simultaneously. To achieve cooling
enhancement, the ESA-induced heat should be less than the heat
ESA cuts from ETU-induced fluorescence quenching. Calculation

shows that the 5I5 manifold is the main source of heating. So the ESA-
induced power density change can be approximately expressed as
ΔPv ≈ β32W32,nrε32Δn3s (4.29)
where Δn3s is the population density change of the 5I5 manifold
after the laser of frequency νe is added to the system, assuming the
pump intensity for GSA is I1 and the pump intensity for ESA is I2. The
population density change of the 5I5 manifold can be approximately
expressed as (see “Appendix”)
(h43 - 1)p11s e C N
2 2
Dn3s ª I2 (4.30)
Ê ˆ Ê hn g W10 ˆ
2 2
2
hn e Á
Ë
Â b3i W3i ˜ Á 2 +
¯ Ë s g I1 ˜¯
i =0
The background absorption of the laser for ESA is

ΔPb = αbI2 (4.31)
For cooling enhancement to occur
ΔPb + ΔPv < 0 (4.32)
needs to be satisfied. By using Eqs. 4.31–4.32, we can obtain
(1 - h43 )x32 e 32 p11s e C N
2 2
ab < a0 = (4.33)
2 2
2
hn e Á
Ë
Â b3i W3i ˜ Á 2 +
¯ Ë s g I1 ˜¯
i =0
where α0 is defined as the critical background absorption coefficient.

Therefore, the co-pumping scheme only improves the cooling
performance of a Ho3+ cooling system with background absorption
lower than α0. In other words, the sample purity is less demanded to
achieve cooling enhancement in a co-pumping cooling system with
a larger α0.
The ETU parameter declines when temperature drops (see
Eq. 4.3); as a result, the upconversion quenching becomes less
problematic. On the other hand, background absorption consistently
brings heat to the sample. Therefore, the co-pumping scheme
may become useless or even harmful to the cooling system as the
temperature drops to a certain value, which can be determined by
using Eq. 4.33. Nevertheless, the co-pumping scheme accelerates
the cooling rate at relatively high temperatures, which is helpful in
shortening the time needed for reaching the minimum attainable

temperature of the cooling system [6].
The cooling power density and cooling efficiency of Ho3+-doped
fluoride crystals via the co-pumping scheme can be simulated by
using the spectral data found in related references. To do that, the
ESA cross section of the 5I5 Æ 5F5 transition is first estimated. In the
frame of Judd–Ofelt theory, the ESA cross section is given by [31]
l m gf
4
Ê ˆ S(l )
se = Á Â A ˜¯ (4.34)
Ú S(l )dl
J
8p c gi Ë J
where λm is the mean absorption wavelength for the 5I5 Æ 5F5

transition. gi and gf are the degeneration coefficients of the initial
and final states, respectively. AJ is the decay rate from 5F5 to each
of the lower manifold (5I4, 5I5, 5I6, 5I7, and 5I8) for an electric dipole
transition. S(λ) is the line shape of the ESA spectrum. Here, the
ESA spectrum shape is roughly assumed to be rectangular, with a
spectrum width of Δλ. The absorption cross section of 5I5 Æ 5F5 can
be approximately expressed as
lm
4
gf Ê ˆ
s e (lm ) ª
8p c Dl gi
Á
Ë
Â A ˜¯ (4.35)
J
J
For Ho3+:YLiF4 crystals, the absorption of the 5I5 Æ 5F5 transition

ranges from 2261 to 2408 nm in wavelength, and the mean
absorption wavelength is 2282 nm [20, 21]. So the ESA absorption
spectrum mainly distributes between 2261 and 2303 nm, and Δλ
is supposed to be 42 nm. The sum of the decay rates from 5F5 to
the lower manifolds is 1798.77 s–1 [20]. On the basis of the above
parametric information, the ESA cross section at 2282 nm is
estimated to be 1.54 × 10–18 cm2.
The spectral data related to 5I7, 5I6, and 5I5 manifolds for a
Ho3+:YLiF4 crystal have been provided in Section 4.2. At 300 K, the
mean wavelength of fluorescence for the 5F5 Æ 5I8 transition is
found to be 649 nm, and the mean wavelengths of fluorescence for
5F Æ 5I , 5F Æ 5I , and 5F Æ 5I transitions are estimated to be
5 7 5 6 5 5
965, 1457, and 2324 nm, respectively [20]. By using Eq. 4.20, the
nonradiative transition rate of the 5F5 Æ 5I5 transition is calculated
to be 15.28 s–1. Other nonradiative transitions are insignificant due
to the large energy gaps. The pump wavelength for GSA and ESA is
selected to be 2070 and 2282 nm, respectively.
Figure 4.7 shows α0 as a function of dopant concentration for

the Ho3+-doped YLiF4 crystal-based cooling system at 300 K. It
shows that α0 increases greatly with the increasing of the dopant
concentration, which can be interpreted from two aspects. For one
thing, increased dopant concentration feedbacks enlarged ESA
absorption (σeCN), so the ratio of the resonant absorption to the
background absorption of the laser for ESA is promoted, and the
system becomes more endurable to the parasitic heat. For another,
the strength of ETU (p11) is proportional to the second order of the
dopant concentration. So, in a densely doped system, upconversion
quenching rather than parasitic absorption is the dominant source
of heating. Besides, the increase in the pump intensity of the laser
for GSA promotes the population density at the 5I7 manifold, and
as a result, the ETU is strengthened and α0 gets larger. It can be
seen from Fig. 4.7 that, by using the co-pumping scheme, cooling
enhancement could be achieved in the Ho3+:YLiF4 crystal with a
dopant concentration of more than 0.4%.
Figure 4.7 Critical background absorption coefficient as a function

of dopant concentration at 300 K. The solid curve is for I1 =
105 W∙m–2, and the dashed curve is for I1 = 107 W∙m–2. The
horizontal dotted line implies available background absorption
in the YLiF4 host (4.0 × 10–4 cm–1). The vertical dotted line
implies the optimum dopant concentration (0.54%) in the
conventional cooling case.
To investigate the temperature dependence of GSA (ESA) and

ETU transition, we define, respectively, the temperature-dependent
coefficients of the GSA (ESA) cross section and ETU parameter as
s g(e) (T )
Q g(e) (T ) = (4.36a)
s g(e) (T0 )
È n(T ) ˘
2
p11 (T )
Q u (T ) = = Í n(T ) ˙ (4.36b)
p11 (T0 ) Î 0 ˚
where T0 = 300 K. Using spectrum data found in Ref. [21], the
temperature-dependent coefficients are plotted in Fig. 4.8. It can be
seen that Θu and Θg decrease with the decreasing of temperature.
When temperature drops to less than 200 K, ETU may become no
longer problematic to the cooling system since the ETU parameter is
tiny. When the temperature approaches 100 K, the small GSA in the
anti-Stokes domain will limit efficient cooling to lower temperatures.
On the contrary, Θe increases with the decreasing of temperature
because the ESA we adopted here is in the Stokes domain. By using
Eqs. 4.3, 4.23, and 4.33, the temperature-dependent α0 is plotted in
Fig. 4.9. It can be seen that when the sample temperature drops to
around 210 K, α0 = αb, which means no cooling enhancement occurs.
At temperatures lower than 210 K, α0 < αb, and the co-pumping
scheme will introduce extra heat to the cooling system (see Eq.
4.32). Therefore, we conclude that the co-pumping for cooling
enhancement only works at relatively high temperatures (for the case
discussed here, T > 210 K). Although it may not be helpful for cooling
at cryogenic temperatures, the co-pumping scheme can shorten the
time needed to reach the minimum attainable temperature of a
Ho3+ cooling system, due to the improved cooling performance at
temperatures of 210–300 K. To estimate the cooling improvement,
next we simulate the cooling power density and cooling efficiency of
the scheme at 300 K.
Generally, the cooling efficiency of an optical refrigeration system
dwindles when the input laser power is increased. As the pump
intensity approaches the saturation of the resonant absorption,
the parasitic absorption will become dominant, and eventually,
the cooling power density is lowered with the declining cooling
efficiency. To obtain the best cooling effect, the intensity of the pump
laser should be deliberately selected. In Ho3+ optical refrigeration
via the co-pumping scheme, the optimal pump intensities of GSA
and ESA for achieving the MCPD appear in pair. Figure 4.10a shows
the attainable MCPD in the Ho3+-doped YLiF4 crystal of various
dopant concentrations at 300 K. The optimal pump intensities for
reaching the MCPD are shown in Fig. 4.10b, and the corresponding
cooling efficiency is shown in Fig. 4.10c. It can be seen that an
MCPD of 7.5 × 107 W∙m–3 could be achieved in the 1.2% Ho3+:YLiF4
crystal. The MCPD in the co-pumping scheme is almost 3 orders of
magnitude higher than that of the conventional laser cooling with
Ho3+:YLiF4 crystals (see Section 4.2). The corresponding cooling
efficiency is about 4.2%, which is several times higher than what
could be obtained via the conventional pumping scheme. As it has
been mentioned, the enhancing effects are due to the cut-down in
the population density of the 5I5 manifold with an ESA of 5I5 Æ 5F5.
Since the decay from the 5F5 manifold is predominantly radiative,
the upconversion quenching is inhibited. Therefore, the dopant
concentration, as well as the pump intensity, could be increased,
resulting in an immense cooling enhancement.
Figure 4.8 Temperature-dependent coefficients of GSA cross section

(solid curve), ESA cross section (dashed curve), and ETU
parameter (dotted curve).
Figure 4.9 Temperature-dependent α0 of the 1% Ho3+:YLiF4 crystal

(I1 = 106 W∙m–2). The dashed line indicates the background
absorption, αb = 4.0 × 10–4 cm–1. The solid point indicates the
critical condition of whether cooling enhancement occurs.
Figure 4.10 At 300 K, (a) the maximum cooling power density (MCPD) as
a function of dopant concentration and (b) the corresponding
optimum pump intensities (OPIs). The solid curve is for GSA,
and the dashed curve is for ESA. (c) The corresponding cooling
efficiency. The background absorption is assumed to be 4.0 ×
10–4 cm–1.
Energy Transfer–Enhanced Laser Cooling in Rare Earth–Co-Doped Solids 133
It is essential to point out that the MCPD of the Ho3+ cooling

system does not always increase with the dopant concentration
because high-order upconversion processes, such as 5I6 Æ 5I8; 5I6
Æ 5S2 (5F4), 5I5 Æ 5I8; 5I5 Æ 5F2 (5F3), and so on, will eventually
emerge when the dopant concentration becomes too high. When
these high-order upconversion processes are strongly involved, the
population evolution of each manifold no longer obeys Eq. 4.22.
The transitions from 5S2, 5F4, 5F3, 5F2, and so on to their adjacent
manifolds are predominantly nonradiative, which lower the cooling
power density, and even turn the system into a heating one. Anyway,
our motivation is to provide a feasible way of realizing efficient
and enhanced cooling in Ho3+-doped solids. In the simulations, the
dopant concentration is restricted within 1.2%. The population
densities of 5I5 and 5I6 manifolds are small enough, so no observable
high-order upconversion processes will involve, and the results are
fairly reliable.
4.4 Energy Transfer–Enhanced Laser Cooling in

Rare Earth–Co-Doped Solids
So far all the discussions of this chapter have been based on the Ho3+-
doped cooling system. In this section, we introduce ETLC in rare
earth ion co-doped systems. The Ho3+, Tm3+:YLiF4 crystal is selected
as the investigated object, in which efficient laser cooling beyond 2
μm could be facilitated. The major ET processes in Ho3+- and Tm3+-
co-doped YLiF4 crystals are illustrated in Fig. 4.11. It is important to
note that the third and the second excited states of Ho3+ and Tm3+
ions could also get involved in the transition processes through the
ET mechanism. The Ho3+ 5I7 Æ 5I8; Ho3+ 5I7 Æ 5I5 and the Tm3+ 3F4 Æ
3H ; Ho3+ 5I Æ 5I ET processes will induce population accumulation
6 7 5
at the 5I5 manifold of Ho3+ ions, and the Tm3+ 3F4 Æ 3H6; Tm3+ 3F4
Æ 3H4 ET processes will induce population accumulation at the 3H4
manifold of Tm3+ ions. Due to that the energy gaps of 5I5–5I6, 5I6–5I7,
3H –3H , and 3H –3F are relatively small, and the following 5I Æ 5I ,
4 5 5 4 5 6
5I Æ 5I , 3H Æ 3H , and 3H Æ3F transitions (dashed arrows) are
6 7 4 5 5 4
highly exothermic [23]. Therefore, inhibiting these ETU processes is
essential for realizing ETLC.
Figure 4.11 Energy level schematic of major Tm3+-Ho3+ ET processes in

a LiYF4 host matrix [23]. The ETU processes of (4), (5), and
(8) result in population accumulation at the third excited
electronic energy level of Ho3+ and Tm3+ ions.
As outcomes of ion–ion interaction, the ET processes depend

on the ion concentrations significantly. Analysis of the fluorescence
spectrum shows that when the concentrations of Ho3+ and Tm3+
ions are restricted within relatively low levels (e.g., in the 0.3%
Ho3+-, 2% Tm3+-doped YLiF4 crystal), the ETU processes mentioned
above could be averted and the Ho3+ 5I7 ´ 5I8; Tm3+ 3H6 ´ 3F4 ET
processes become dominant [32]. In this case, the Ho-Tm ETLC that is
shown schematically in Fig. 4.12 could be realized with proper laser
excitation. The incident laser photons are tuned to be resonant for
the transitions from the higher 5I8 to the lower 5I7 manifolds (vertical
solid arrow on the left). The exited Ho3+ ions relax thermally among
the manifolds and then decay by emitting fluorescence photons at a
mean frequency of ν1 (vertical dashed arrow on the left). For each
excitation and de-excitation cycle, an average thermal vibration
energy of hν1 – hν is removed from the host material with extraction
of the fluorescence photon. This is the case of conventional ASFC
usually discussed.
Meanwhile, part of the ground-state Tm3+ ions gets sensitized
by absorbing the energy transferred from Ho3+ excitations and the
surrounding phonon energy (phonon-assisted ET, double-head
arrow) [28] and then decays by emitting fluorescence photons at a
mean frequency of ν2 (vertical dashed arrow on the right). Since the
pump laser is red-edge- tuned for Ho3+ ion excitation, the absorption
of Tm3+ ions is negligible (see Fig. 4.2d). For each excitation and de-
excitation cycle of Tm3+ ions, an extra thermal vibration energy of hν2
– hν1 will be released. According to ET theory, the density population
dynamic of the electronic energy levels of the above Ho-Tm cooling
system can be described by the following equations:
dn1 Pr
= - he,1W1,r n1 - W1,nr n1 - p13 n1 n3 + p02 n0 n2 (4.37a)
dt hn
dn2
= -he,2W2 ,r n2 - W2 ,nr n2 + p13 n1 n3 - p02 n0 n2 (4.37b)
dt
n0 + n1 = CHoN (4.37c)
n2 + n3 = CTmN (4.37d)
where ni (i = 0, 1, 2, 3) is the population of the ith level, as shown
in Fig. 4.12. Pr is the resonantly absorbed power density at pump
frequency ν. Wj,r and Wj,nr are the radiative decay rate and the
nonradiative decay rate, respectively (j = 1 for the Ho3+ 5I7 Æ 5I8
transition and j = 2 for the Tm3+ 3F4 Æ 3H6 transition). p13 and p02
are the temperature-dependent macroscopic ET rates [33]. CHo
and CTm are the percentage concentrations of Ho3+ and Tm3+ ions,
respectively.
Figure 4.12 Schematic of Ho-Tm ETLC.
The net power density transferred to the sample is obtained by

using the energy conservation law
Pnet = –ηe,1W1,rn1,shv1 – ηe,2W2,rn2,shv2 + Pr + Pb (4.38)
where ni,s is the steady-state population of the ith energy level. The
first term on the right-hand side of Eq. 4.38 represents the energy
that is taken out of the host by the conventional ASFC mechanism

of Ho3+ ions, and the second term takes into account the cooling
contribution of ET processes. The cooling efficiency, which is defined
as the ratio of the cooling output to the total absorption, is expressed
as
- Pnet he,1W1,r n1 ,sn 1 + he,2W2,r n2,sn 2
hc = = habs - 1 (4.39)
Pr + Pb (W1 n1,s + W2 n2,s )n
where W1(2) is the total effective decay rate that includes both the
radiative decay rate with consideration of florescence reabsorption
and nonradiative decay rate. Substituting the analytical results of ni,s
into Eq. 4.39, a symmetrical expression is obtained,
Ê hext,1n 1 + hext,2n 2 hext,1n 1 - hext,2n 2 ˆ
hc = habs Á + ˜¯ - 1 (4.40)
Ë 2n 2n c
where ηext,1(2) is the external quantum efficiency for the Ho3+ 5I7 Æ
5I (Tm3+ 3F Æ 3H ) transition, and
8 4 6
vc = Pr/hA (4.41)
with
È 4 p13C TmW2 N ˆ ˘
1/2
Ê
2
Pr Pr
A=B-Í +ÁB - ˜ ˙ (4.42)
ÍÎ p02 - p13 hn Ë hn ¯ ˙˚
and
( p13C TmW2 + p02C HoW1 )N + W1W2
B= (4.43)
p02 - p13
The appearance of variable νc in the cooling efficiency formula

(Eq. 4.40) of ETLC is a result of the ET mechanism acting on the
co-doped cooling system. As an important and unique parameter
of ETLC, νc determines the cooling efficiency of a co-doped cooling
system. It can be seen from Eqs. 4.41–4.43 that νc, which will be
referred next as the characteristic frequency, lies on four pairs of
parameters: the inputs (Pr and ν), the total effective transition rates
(W1 and W2) that can be treated as constants, the temperature-
dependent macroscopic ET rates (p13 and p02), and the percentage
concentrations (CHo and CTm). The merit of Eq. 4.40 is that knowing
the basic property of the characteristic frequency of an ETLC
system, one will quickly draw a qualitative conclusion of the cooling
efficiency from the expression. For instance, if the concentration of
Tm3+ ion equals zero, that is, the Ho3+–Tm3+ ET processes do not
exist (p13 = 0), the value of the characteristic frequency then equals
that of the pumping frequency (νc = ν). Equation 4.40 can be reduced
to
n1
hc n c =n
= hext,1habs - 1 (4.44)
n
This is exactly the cooling efficiency formula of the conventional
ASFC system. As a general expression of cooling efficiency for the
Ho-Tm ETLC system, Eq. 4.40 shows that the larger the characteristic
frequency of an ETLC system is, the larger the corresponding cooling
efficiency could be. The above analysis shows that the ETLC cooling
scheme is more efficient than the conventional ASFC scheme in
principle, so the right-hand side of Eq. 4.40 should be larger than
that of Eq. 4.44, indicating that νc > ν. Indeed, for the 0.3% Ho3+, 2%
Tm3+:YLiF4 crystal sample, the calculated result on the characteristic
frequency (with transition rates and macroscopic ET rates near 300
K found in Refs. [23, 34]) shows that the value of the characteristic
frequency is on the order of 2ν.
Next, we predict the cooling performance of an ETLC system
on the basis of spectrum data and material parameters of Ho3+,
Tm3+:YLiF4 crystals. The energy gap of Tm3+ 3F4–3H6 is 5810 cm–1
[23]. By using Eq. 4.20, the nonradiative decay rate of the Tm3+
3F Æ 3H transition in the YLiF crystal is calculated as 0.10 s–1.
4 6 4
The radiative decay rate of this transition is 66.67 s–1 [23]. The
external quantum efficiencies of Ho3+ 5I7 Æ 5I8 and Tm3+ 3F4 Æ 3H6
transitions in the YLiF4 host are calculated to be 0.9949 and 0.9985,
respectively. The average fluorescence wavelength for the Tm3+ 3F4
Æ 3H6 transition is found to be 1807 nm (see Fig. 4.2d).
Figure 4.13a shows the cooling efficiencies of the 0.3% Ho, 2%
Tm YLiF4 and the 0.3% Ho YLiF4 crystal samples as functions of
resonant absorption at a pump wavelength of 2070 nm. To point
out, since the dopant concentration of Ho3+ ions is relatively low, for
the sake of brevity, the model M1 (see Section 4.2) has been applied
in simulating Ho ASFC. As the resonant absorption increases,
the absorption efficiency will decrease; as a result, the cooling
efficiency of Ho ASFC drops. On the contrary, the cooling efficiency
of Ho-Tm ETLC first slightly rises with the increasing of the resonant
absorption, which is beneficial for boosting the cooling power density.
In essence, as more Ho3+ ions are pumped to the excited state with
the increased absorption, the average distance between an excited-
state Ho3+ ion and a ground-state Tm3+ ion shrinks. Consequently,
the Ho3+–Tm3+ ET is accelerated and more Tm3+ ions participate in
the cooling cycle. At a resonant absorption of around 0.8 × 108 W/
m3, the 0.3% Ho, 2% Tm ETLC system reaches its maximum cooling
efficiency (5.08%). However, the cooling efficiency for the 0.3% Ho
ASFC system at a resonant absorption of 0.8 × 108 W/m3 is only
about 0.93%. The improving in the cooling efficiency by applying the
ET mechanism to a solid-state cooling system is quite remarkable.
Moreover, the maximum resonant absorption for cooling (ηc > 0) of
the 0.3% Ho, 2% Tm ETLC system is roughly two times of that of the
0.3% Ho ASFC system, indicating that the co-doped cooling scheme
could profoundly enhance the cooling power density. It can be seen
from Fig. 4.13b, at a resonant absorption of about 0.7 × 108 W/m3,
the MCPD for the 0.3% Ho ASFC system is 0.8 × 106 W/m3, whereas
at a resonant absorption of about 1.8 × 108 W/m3, the MCPD for the
0.3% Ho, 2% Tm ETLC system is 8 × 106 W/m3.
Figure 4.13 The cooling efficiencies (a) and the cooling power densities (b)
of Ho-Tm ETLC and Ho ASFC in the YLiF4 crystal as functions
of resonant absorption near 300 K. For the 0.3% Ho, 2% Tm:
YLiF4 crystal sample, the ET parameters are estimated as p13
= 6 × 10–24 cm3∙μs–1 and p02 = 8 × 10–23 cm3∙μs–1 [34]. The
background absorption is assumed to be 4.0 × 10–4 cm–1.
The ET-induced cooling enhancement varies with the sample

temperature since the macroscopic ET rates contained in the
characteristic frequency are temperature dependent. To make a
clear statement, we consider the low excitation cases, in which less
than 5% of the Ho3+ ions are pumped to the excited state for the
sample mentioned above. B that is defined by Eq. 4.43 will be more
than 20 times larger than Pr/hν. The right-hand side of Eq. 4.42 can
then be expanded as a power series of Pr/hν. Neglecting the terms
that are higher than second order, Eq. 4.42 can be approximately
expressed as
Pr p02C HoW1 - p13C TmW2
Aª (4.45)
hn p02C HoW1 + p13C TmW2
where the condition (p02CHoW1 – p13CTmW2)N >> W1W2 has been

used in the simplification. Substituting Eq. 4.45 into Eq. 4.40, we
obtain the cooling efficiency of the ETLC system in the low excitation
condition,
Ê n1 ˆ hext,2n 2 - hext,1n 1
hETA = Á habs hext,1 - 1˜ + Ghabs (4.46)
Ë n ¯ n
where the enhancing factor that marks the ET-induced cooling
enhancement is expressed as
-1
Ê W1 C Ho p02 ˆ
G (T , C Tm C Ho ) = Á 1 + ˜ (4.47)
Ë W2 C Tm p13 ¯
It can be seen clearly that the terms in the parentheses of

Eq. 4.46 represent the conventional ASFC of Ho3+ ions, and the last
term on the right-hand side accounts for the cooling enhancement
induced by the ET mechanism. Equation 4.47 shows that to
increase the concentration ratio of the Tm3+–Ho3+ ion will bring a
larger enhancement in the cooling efficiency. This feature of ETLC
is expected since an increased Tm3+ concentration promotes the
interacting probability of the ground-state Tm3+ ions with the
excited-state Ho3+ ions. It needs to be noted that both of the Ho3+ and
Tm3+ ions should be restricted within relatively low concentration
levels to avoid ETU-induced heating (see Section 4.2), whereas to
break through the concentration limits and increase the cooling
power density, it is possible to adopt alternative low–host phonon
materials (e.g., KPb2Cl5 crystals), since the multiphonon transitions
may be hindered [18].
The ratio of p13 to p02, which is usually referred to as the
temperature-dependent equilibrium constant, governs the steady-
state energy assignment of a Ho-Tm cooling system. By using the
approximate expression of the temperature-dependent equilibrium

constant derived by Walsh et al. [34], the enhancing factor as a
function of temperature is depicted in Fig. 4.14. It can be seen that
the enhancing factor decreases with the decreasing of the temperate.
Actually, as the temperature drops, more excited-state Ho3+ ions
will occupy the lower manifolds of the electronic energy level due
to the nature of Boltzmann distribution. The enlarged average
energy difference between the excited electronic energy levels of
the Tm3+ and Ho3+ ions weakens the Ho3+–Tm3+ ET processes, and
consequently the enhancing factor decreases. Nevertheless, the
ET mechanism assists the cooling processes consistently to the
minimum achievable temperature of the cooling system since the
value of the enhancing factor is always positive. So the timescale
for solid-state laser cooling could be shortened by applying the co-
doped cooling scheme.
Figure 4.14 The enhancing factor as a function of temperature in a low

excitation condition.
It is worth mentioning that the co-doped cooling scheme can also

be applied to rare earth–co-doped materials other than the Ho3+,
Tm3+:YLiF4 crystals, given that the selected dopants share a similar
electronic energy level structure so that ET processes can efficiently
occur. For instance, the Tm3+, Er3+-co-doped materials are favorable
options for ETLC, since the first excited states (3F4 and 4I13/2) of
Tm3+ and Er3+ ions are close enough in energy. It is important to
Conclusions 141
note that the host material should enable ASFC of each dopant in
the first place, and the doping concentrations should be restricted
within relatively low levels so that the deleterious ETU processes are
averted. Another couple of dopants for ETLC could be Er3+ and Yb3+,
since the Er3+ 4I11/2 and Yb3+ 2I5/2 states are very close in energy.
However, the drawback of Er-Yb ETLC is that the intermediate Er3+
11/2 Æ I13/2 transition usually produces a large amount of heat that
4I 4
may impede the cooling processes. Applying a low–phonon energy

(which is small enough compared to the energy gap of Er3+ 4I11/2
– 4I13/2, so the nonradiative transition is negligible) host material
could solve the heating issue.
4.5 Conclusions
This chapter theoretically investigates optical refrigeration in
Ho3+:YLiF4 and Ho3+, Tm3+:YLiF4 crystals. Due to the small energy
mismatch between transitions 5I7 Æ 5F8 and 5I7 Æ 5I5, ETU occurs
when the 5I5 Ho3+ ion is pumped in the anti-Stokes domain. Therefore,
a general cooling model with consideration of upconversion is
established to precisely describe laser cooling of Ho3+-doped
fluoride crystals. Meanwhile, a guide on selecting a host material for
optical refrigeration with multimanifold rare earth ions is provided.
Simulation results show that the multiphonon transitions from the
second and third excited states of Ho3+ ions block effective cooling.
To this end, a co-pumping scheme is presented to solve the ETU-
induced fluorescence quenching issue. An extra laser is added to the
cooling system to pump 5I5 Ho3+ ions to a higher excited state (5F5).
The beauty of applying the co-pumping scheme is that the 5F5 Ho3+
ions scarcely transit to the lower nearby state and mainly de-excite
radiatively. So, the large amount of heat produced by nonradiative
5I Æ 5I and 5I Æ 5I transitions in the conventional anti-Stokes
5 6 6 7
cooling case can be avoided, and both cooling efficiency and cooling
power can be prompted. Besides, there is another way of improving
the cooling performance of Ho3+ optical refrigeration: co-doping
Tm3+ and Ho3+ ions in the host. The Ho3+ and Tm3+ ions share a
similar electronic energy level structure, and their first excited
states are close in energy. As a result, ET from Ho3+ to Tm3+ ions
arises when Ho3+, Tm3+-co-doped fluoride crystals are pumped at
around 2.07 μm. Since the energy gap of Tm3+ 3F4 - 3H6 is wider than
that of Ho3+ 5I7 - 5I8, host phonons will be annihilated when ET from
Ho3+ to Tm3+ ions takes place, bringing enhancement to the cooling
effect. However, one weak point of ETLC is that both of the two
dopants should be restricted within relatively low concentration
levels to avoid ETU-induced heating.
To conclude, ETU-induced fluorescence quenching is the main
obstacle to achieve net cooling in Ho3+-doped solids. Nevertheless,
by adopting the modified cooling schemes and the optimization
procedures presented in this chapter, it is possible to overcome
the ETU-induced heating and realize efficient optical refrigeration
around 2.07 μm.
Acknowledgments
The authors acknowledge financial support of the Specialized
Research Fund for the Program for New Century Excellent Talents
in University (grant no. NCET-11-269), the National Natural Science
Foundation of China (grant nos. 11204048, 61275138, 10804022,
60878016, and 61378027), the Fundamental Research Funds for
the Central Universities, the Special Financial Grant from the China
Postdoctoral Science Foundation (grant no. 2013T60345), and the
111 project to the Harbin Engineering University (grant no. B13015).
Appendix
For simplicity, the pump intensities for both GSA and ESA are
assumed to be low, so the population densities of 5I6, 5I5, and 5F5
manifolds are far less than the 5I7 manifold. In the conventional
cooling case, the steady-state population density (nis1, i = 0, 1, 2, 3)
of each manifold can be obtained from the following equations:
-( b30W30 + b31W31 + b32W32 )n3s1 + p11 (C )n1s1 = 0 (A.1a)
2
I1
s g (n0 s 1 - n1 s 1 ) - W10 n1 s 1 ª 0 (A.1b)
hn g
n0s1 + n1s1 ≈ CN. (A.1c)

The approximate expression for the population density of the 5I5
manifold in the conventional pumping scheme reads
References 143
2 2
p11 C N
n3s1 ª (A.2)
2 2
ÁË Â b3i W3i ˜ Á 2 +
¯Ë s I ˜¯
i =0 g 1
In the cooling case via the co-pumping scheme, the equations

governing the steady-state population density of 5I7 and 5I8 (n1s2 and
n0s2) are identical to Eqs. A.1b and A.1c. But the equations governing
the steady-state population density of 5I5 and 5F5 (n3s2 and n4s2) are
written as
I2 Ê 3
ˆ
s e n3s2
hn e
-Á
Ë
Âb 4i
W4 i ˜ n4 s 2 ª 0 (A.3)
¯
i =0
I2 Ê 2
ˆ
-s e n3s2
hn e
-Á
Ë
Âb 3i
W3i ˜ n3s2 + b 43W43 n4s2 + p11 n1s2 ª 0 (A.4)
¯
2
i =0
The approximate expression for the population density of the 5I5

manifold in the co-pumping scheme reads
2 2
p11 C N
n3s2 ª (A.5)
È s e I2 Ê ˆ ˘ Ê hn g W10 ˆ
2 2
Í(1 - h43 ) +
hn e ÁË i = 0
b3i W3i ˜ ˙ Á 2 +
¯ ˚Ë
Â
s g I1 ˜¯
Î
The population density change of the 5I5 manifold induced by
adding the laser for ESA to the system is
-2
Ê hn g W10 ˆ
Dn3s = n3s2 - n3s1 = p11 C N ÁË 2 + s I ˜¯
2 2
g 1
(A.6)
ÏÔ È s e I2
2
˘
-1
Ê 2
ˆ ¸Ô
-1
¥ Ì Í(1 - h43 )
hn e
+ Â b3i W3i ˙ -Á
Ë
Â b3i W3i ˜ ˝
¯ Ô˛
ÔÓ Î i =0 ˚ i =0
Since I2 is small, the right-hand side of Eq. A.6 can be expanded

as a power series of (1–η43)σeI2(hνe)–1, and the result formulated in
Eq. 4.30 is obtained.
References

Nature, 377, pp. 500–503.
2. Epstein, R. I., Sheik-Bahae, M. (2009). Optical Refrigeration (Wiley-

VCH, Weinheim).
3. Sheik-Bahaeand, M., Epstein, R. I. (2009). Laser cooling of solids, Laser
Photonics Rev., 3, pp. 67–83.
Phys., 73, p. 086501.
5. Seletskiy, D. V., Melgaard, S. D., Bigotta, S., Lieto, A. D., Tonelli, M., Sheik-
6. Seletskiy, D. V., Melgaard, S. D., Epstein, R. I., Lieto, A. D., Tonelli, M.,
Sheik-Bahae, M. (2011). Local laser cooling of Yb:YLF to 110 K, Opt.
Express, 19, pp. 18229–18236.
7. Melgaard, S. D., Seletskiy, D. V., Lieto, A. D., Tonelli, M., Sheik-Bahae,
1588–1590.
8. Melgaard, S., Albrecht, A. R., Hehlen, M., Seletskiy, D., Sheik-Bahae, M.
(2014). Optical refrigeration cools below 100K, in CLEO: 2014, OSA
Technical Digest (online) (Optical Society of America), paper FTh4D.4.
9. Garcia-Adeva, A. J., Balda, R., Fernandez, J. (2009). Upconversion
cooling of Er-doped low-phonon fluorescent solids, Phys. Rev. B, 79, p.
033110.
10. Nemova, G., Kashyap, R. (2010). Laser cooling of Er3+-doped solids,
Opt. Commun., 283, pp. 3736–3739.
11. Dong, G. Z., Zhang, X. L., Li, L. (2013). Energy transfer enhanced laser
cooling in Ho3+ and Tm3+ codoped lithium yttrium fluoride, J. Opt. Soc.
Am. B, 30, pp. 939–944.
12. Dong, G. Z., Zhang, X. L. (2015). Efficient and enhanced optical
refrigeration beyond 2 μm in Ho3+ doped solids via co-pumping
scheme: erratum, J. Opt. Soc. Am. B, 32, pp. 2248–2249.
E. (2000). Observation of anti-Stokes fluorescence cooling in thulium-
cooling in bulk erbium-doped materials, Phys. Rev. Lett., 97, p. 033001.
R. I. (2008). Anti-Stokes luminescence cooling of Tm3+doped BaY2F8,
Opt. Express, 16, pp. 1704–1710.
References 145
16. Condon, N. J., Bowman, S. R., O’Connor, S., Quimby, R. S., Mungan, C.
5472.
17. Bowman, S. R., O’Connor, S., Condon, N. J., Friebele, E. J., Kim, W., Shaw,
B., Quimby, R. S. (2013). Non-radiative decay of holmium-doped laser
materials, Proc. SPIE, 8638, p. 863803.
18. Hehlen, M. P. (2009). Novel materials for laser refrigeration, Proc. SPIE,
7228, p. 72280E.
19. Dong, G. Z., Zhang, X. L. (2015). Role of upconversion in optical
refrigeration: a theoretical study of laser cooling with Ho3+ doped
fluoride crystals: erratum, J. Opt. Soc. Am. B, 32, pp. 2046–2047.
20. Walsh, B. M., Barnes, N. P., Bartolo, B. D. (1998). Branching ratios, cross
sections, and radiative lifetimes of rare earth ions in solids: application
to Tm3+ and Ho3+ ions in LiYF4, J. Appl. Phys., 83, pp. 2772–2787.
21. Walsh, B. M., Grew, G. W., Barnes, N. P. (2005). Energy levels and
intensity parameters of Ho3+ ions in GdLiF4, YLiF4 and LuLiF4, J. Phys.:
Condens. Mater., 17, pp. 7643–7665.
22. Barnes, N. P., Walsh, B. M. (2003). Ho:Ho upconversion: applications to
Ho lasers, J. Opt. Soc. Am. B, 20, pp. 1212–1219.
23. Walsh, B. M., Barnes, N. P., Petros, M., Yu, J., Singh, U. N. (2004).
Spectroscopy and modeling of solid state lanthanide lasers: application
to trivalent Tm3+ and Ho3+ in YLiF4 and LuLiF4, J. Appl. Phys., 95, pp.
3255–3271.
24. Hoyt, C. W., Hasselbeck, M. P., Sheik-Bahae, M., Epstein, R. I., Greenfield,
25. Dexter, D. L. (1952). A theory of sensitized luminescence in solids, J.
Chem. Phys., 21, pp. 836–850.
26. Rapaport, A., Milliez, J., Bass, M., Cassanho, A., Jenssen, H. (2005).
Numerical model of the temperature dependence of the up-conversion
efficiency of fluoride crystals codoped with ytterbium and thulium, J.
Appl. Phys., 97, p. 123527.
27. van Dijk, J. M. F., Schuurmans, M. F. H. (1983). On the nonradiative and
radiative decay rates and a modified exponential energy gap law for
4f-4f transitions in rare-earth ions, J. Chem. Phys., 78, pp. 5317–5323.
28. Powell, R. C. (1998). Physics of Solid-State Laser Materials (Springer-
Verlag, New York).
29. Lee, T. H., Heo, J. (2005). 1.6 μm emission and gain properties of Ho3+
in selenide and chalcohalide glasses, J. Appl. Phys., 98, p. 113510.
30. Payne, S. A., Chase, L. L., Smith, L. K., Kway, W. L., Krupke, W. F. (1992).
Infrared cross section measurements for crystals doped with Er3+,
Tm3+, and Ho3+, IEEE J. Quantum Electron., 28, pp. 2619–2630.
31. Librantz, A. F. H., Jackson, S. D., Gomes, L., Ribeiro, S. J. L., Messaddeq,
Y. (2008). Pump excited state absorption in holmium-doped fluoride
glass, J. Appl. Phys., 103, p. 023105.
32. Faiconieri, M., Salvetti, G. (1994). Effects of co-dopant concentrations
and excitation conditions on the 2 μm fluorescence dynamics in Tm,
Ho: YLF crystals. Appl. Phys. A, 59, pp. 253–258.
33. Walsh, B. M., Barnes, N. P., Bartolo, B. D. (1997). On the distribution of
energy between the Tm 3F4 and Ho 5I7 manifolds in Tm-sensitized Ho
luminescence, J. Lumin., 75, pp. 89–98.
34. Walsh, B. M., Barnes, N. P., Bartolo, B. D. (2000). The temperature
dependence of energy transfer between the Tm 3F4 and Ho 5I7
manifolds of Tm-sensitized Ho luminescence in YAG and YLF, J. Lumin.,
90, pp. 39–48.
Chapter 5
Optically Cooled Lasers
Steven R. Bowman
Laser Physics Branch, Naval Research Laboratory, Washington, DC 20375-5320, USA
steven.bowman@nrl.navy.mil
Since the invention of the laser, engineers have searched for new
ways to increase the laser’s power and brightness. These efforts
have largely focused on improved ways to remove the waste heat
that is generated in the lasing medium. This chapter reviews a novel
approach to the problem of laser power scaling: minimizing the
waste heat through the incorporation of anti-Stokes fluorescence
cooling. This approach has been shown to significantly reduce
heat generation in carefully controlled laser systems. This chapter
reviews the theory and practice of this novel mode of laser operation.
5.1 Introduction
One of the more useful properties of lasers is their ability to generate
extremely high power densities. Take, for example, a commercial
fiber laser operating continuously with an average power of 1 kW.

www.panstanford.com
148 Optically Cooled Lasers
Inside the small core of the fiber the beam propagates at intensities
as high as 1 GW/cm2. For comparison, this intensity is 10 billion
times higher than terrestrial sunshine. Even the hardest metals
are instantly vaporized when placed near the focus of this beam.
Yet the laser that generates these intensities is expected to operate
reliably for years. The challenge for the laser designers is to cleanly
manipulate such high intensities without damaging the laser
medium.
What happens to a material when it is exposed to a high-intensity
laser depends principally on the amount of power absorbed. Most
materials convert absorbed laser power directly into heat. When the
absorption is described by a Beer’s law coefficient of α and the laser
intensity is IL, then the absorbed power density is simply αIL. So any
material with significant absorption placed in the beam described
above would see GW/cm3 heat loads and be instantly vaporized.
Even the most transparent materials, such as a pure silica fiber,
have weak background absorption with coefficients of order of 10–6
cm–1. So propagation of the 1 kW laser beam described above will
still generate 1 kW/cm3 of waste heat. The core of an optical fiber
survives this extreme heat loading by spreading the thermal load
over a large surface area. Less transparent materials, such as a laser
gain medium, can see even higher thermal loads. As a result, thermal
loading remains the primary obstacle to scaling up the average
power of most laser systems.
5.2 Laser Thermal Management

The problem of thermal management of the laser is as old as the
laser itself. Many groups have contributed to the large body of work
and tremendous progress on the problem. This section is intended
only as only a brief summary of the major elements of the field. We
will start by examining the source and magnitude of the problem,
restricting the discussion to optically pumped lasers.
Earlier it was stated that the power density of heat generated
in most materials with absorption coefficient a by the passage of a
laser beam with intensity IL would be aIL. But this is not generally
true for laser media. Laser media are designed to reemit absorbed
light with high efficiency. The amount of heat generated in a laser
Laser Thermal Management 149
depends on the fraction of the pump power that is not reradiated as

well as the spectra of the emission. The reradiated light will normally
be red-shifted relative to the pump. So even if the excitation and
lasing processes were 100% efficient, every laser photon emitted
will contribute a fraction of its energy QD = (1 – lP/lL) to heating the
laser material. Here lP and lL are the pump and laser wavelengths
and QD is the laser quantum defect (QD).
Heat generated in the laser’s interior will produce a temperature
rise in the center of the gain medium relative to its edge. This
temperature rise produces several detrimental effects on the laser
beam propagating through it. Firstly, it can reduce the lifetime and
increase the background absorption of the lasing ions. These effects
combine to reduce net gain. The temperature profile will also modify
the refractive index and thickness of the medium distorting the
laser beam. The strain generated by the temperature gradient also
perturbs the index and can result in depolarization of the beam. If
the strain becomes large enough, the laser medium will crack and
fail completely. Of course if the temperature rises too high, the laser
medium will simply melt or burn. Usually beam quality degradation
limits average power well below thermal damage.
To illustrate the detrimental impact of thermal loading consider
one of the most common solid-state lasers: neodymium-doped
yttrium aluminum garnet (Nd:YAG) operating at 1064 nm. YAG’s
excellent thermomechanical properties have made it the most
popular material for high-average-power solid-state lasers. Nd:YAG
is a strong 1 µm emitter with a quantum efficiency near unity.
Experiments have shown that when pumped near the optimum
absorption at 807 nm, a continuous-wave (cw) Nd:YAG laser converts
25% of the absorbed pump into heat. This is largely the result of the
23% QD between the pump and laser photon energies.
In a simple end-pumped cw Nd:YAG laser, 10 W of pump powers
was shown to generate as much as 7 kW/cm3 of heating and axial
temperature rise up to 37 K [56]. This produced a 14-diopter
thermal lens in the YAG crystal with a strong spherical aberration.
Further, the thermal loading will roughly double when the crystal
is not lasing. This is due to strong upconversion and multiphonon
quenching. Still, laser systems based on Nd:YAG hold the record
for average power (15 kW) with good beam quality [48]. This is a
testament to decades of innovations and refinements by the solid-
state laser community. Nevertheless, heat generation in Nd:YAG and

other conventional bulk laser materials limits their maximum useful
average power.
Recently laser architectures based on silica fibers have also made
great advances in average power performance. Fiber lasers deal
with heat dissipation by virtue of their high aspect ratios, typically
of the order of a million to one. This allows them to greatly spread
the thermal loading on the laser medium. Also, wave guiding in the
fiber’s core makes the laser beam much less susceptible to thermo-
optic distortions. Fiber lasers based on the 1 µm transition in
ytterbium have demonstrated the highest average powers. Currently
the highest-power systems produce up to 10 kW with good beam
quality [31]. Nevertheless, at the kilowatt average power level, fiber
lasers also suffer from internal heating. Longitudinal temperature
gradients in the fiber have been shown to generate Raleigh scattering,
which destabilizes the transverse mode and spoils the beam quality
[61, 64]. As with bulk laser, thermal loading currently limits the
maximum average power from a single-fiber laser.
The current best approach to further power scaling is to combine
the beams from many parallel lasers. Beam combination can be
accomplished through precise control of the phases, wave vectors,
or wavelengths of an ensemble of lasers. Each of these approaches
has been proven to increase useful power but at the expense of
dramatically increasing the complexity of the overall system. Power
scaling with this approach rapidly runs into practical limits on the
amount of hardware acceptable for the application [58].
Many other strategies have been explored to help manage
the thermal loading in high-power lasers. Slab and disk laser
geometries that expedite heat dissipation have been fully explored
[1, 3, 19, 20, 21, 25, 33, 34, 39, 41]. Wavefront correction techniques
such as nonlinear phase conjugation have also been well studied
[17]. Pulsed laser approaches have been demonstrated that allow
transient heat build-up [2]. New laser materials with higher thermal
conductivities are still an active research area [54, 55, 59, 63]. In
the early days of lamp-pumped lasers, many systems operated at
cryogenic temperatures. Recently cryogenic operation has been
revisited to improve material thermo-optic parameters [26]. Finally,
laser schemes that minimize the laser QD have been developed in
order to reduce heating [16, 18, 43, 55].
Theory of Low–Quantum Defect Laser Materials 151
The remainder of this chapter explores this last approach:

low–quantum defect lasers (LQDL). Specifically we consider the
possibility of anti-Stokes fluorescence cooling that can arise in
LQDL schemes. Anti-Stokes emission reduces the overall QD further
and removes waste heat from the laser medium. The special case
of radiation-balanced lasing will be considered in detail. This
occurs when fluorescence cooling just cancels out other heating
mechanisms. Unique among all the power scaling approaches, a
radiation-balanced laser (RBL) can theoretically have zero thermal
loading. So an RBL system could potentially break through the
average power barrier that has limited the other approaches.
5.3 Theory of Low–Quantum Defect Laser

Materials
This section develops a theory for solid-state lasers that operate
with minimal excess heating from the laser QD. The goal is to develop
a realistic model that can be used to explore the fundamental
limitations of laser performance. The approach is to simplify the
analysis by just considering the effects inside the lasing medium. By
tracking the local power density in the laser medium insights can be
obtained about radiative efficiency and waste heat generation. This
analysis applies to the most common type of high-average-power
lasers: the steady-state quasi-three-level systems.
Several previous studies have developed models for quasi-three-
level laser systems [4, 6, 27, 65]. These models have largely focused
on predicting gain and power for specific laser architectures, such as
an end-pumped resonator or amplifier. As a result, they have focused
on the impact of beam profiles and laser geometry. To simplify the
analysis these models have largely ignored the contribution of
fluorescence to power flow. We will show that this can lead to an
underestimate of the gain and an overestimate of the heat loading.
The analysis presented below is intended to augment the previous
resonant-pumping studies. It also extends and generalizes previous
analysis of RBLs [14, 46, 50].
The model present below predicts the local performance
of a quasi-three-level laser material. After developing the basic
definitions and relations in a lossless idealize material, the analysis
will be extended to include the most common processes: quenching,

background absorption, and fluorescence reabsorption. Several
possible optimization approaches will be considered and compared.
The well-characterized laser material Yb:YAG will be used to illustrate
these relations and optimization strategies. The techniques used in
this analysis can be readily extended to specific gain geometries and
beam profiles.
5.3.1 Lossless Laser Materials

LQDLs are typically based on a transition between an energy level
just above the ground state and the first excited manifold of a dopant
ion. These lasers are often referred to as quasi-three-level systems,
but they actually consist of only two well-separated energy bands
made up of several nondegenerate levels. Each of these energy
manifolds maintains a thermal equilibrium with the host matrix
through electron–phonon scattering. The laser transition terminates
in a level just high enough above the ground state to have a very
small thermal population. So unlike Nd:YAG, lasers based on ions
like ytterbium can avoid detrimental energy pooling in the lower
level, while still maintaining a low QD.
We begin our analysis by considering an ideal laser system with
only radiative losses and no radiative trapping. Here the power
flow of the fluorescent field will be included, unlike conventional
laser analysis. After deriving some basic results, the model will be
extended to handle more realistic cases.
Assume an isotropic laser medium that is doped with active ions
at a density of NT. Optical transitions are presumed to occur between
lower and upper manifolds, giving rise to overlapping, thermally
broadened absorption and emission spectra. These spectra are
characterized by effective absorption and emission cross sections
labeled sA(l,T) and sE(l,T), respectively, at a wavelength l and a
temperature T. It will be useful to define a fractional cross section as
s A ( l ,T )
b ( l ,T ) ∫ . (5.1)
s A ( l ,T ) + s E ( l ,T )
Ion densities in the lowest-energy manifold, N1, are excited to the

upper-energy manifold, N2, by a pump field at wavelength lP and
irradiance IP. Ions in the upper manifold radiate with a lifetime tR or
are stimulated to emit decay by the laser field at wavelength lL and
irradiance IL. For simplicity, we assume that the pump and laser fields
are directional, monochromatic, and polarization independent. This
is not always the case, but the model is easily modified to include
broader intensities and nonisotropic media. Our derivation also
assumes that ionic dephasing is very rapid compared to tR, allowing
us to consider only incoherent excitation. With these definitions and
mass conservation, we can write the local rate equation as
dN2 IPlPs A (lP ,T )
= ◊ ÈÎ NT - N2 b(lP ,T )˘˚
dt hc
I l s ( l ,T )
+ L L A L ◊ ÈÎ NT - N2 b(lL ,T )˘˚ - N2 t R .(5.2)
hc
The first term on the right describes the pump rate, the second
term describes the laser rate, and the last term describes the total
spontaneous decay rate of the upper manifold. Applying energy
conservation to this laser system, we recognize that the difference
between the absorbed power and the emitted power must be the
local heat generation rate, H. With the definitions above the power
density of waste heat will be
hc N2
H = IPs P ◊ È NT - ˘+I ◊ È NT - ˘-
N2 N2
Ls L . (5.3)
ÎÍ b P ˚˙ ÎÍ b L ˚˙ lF t R
Note that the heat generation depends on the mean fluorescence

wavelength, lF, defined as the average over the emission band:
lF ∫
Ú band
l ◊ IF ( l ) d l
. (5.4)
Ú
band
IF ( l ) d l
We will simplify the derivation using the shorthand notations

s P ∫ s A ( lP , T ), s L ∫ s A ( lL , T ),
bP bL
b P ∫ b ( lP , T ), b L ∫ b ( lL , T ), and b PL ∫ . (5.5)
bP - bL
The rate equation above makes it clear that the appropriate

saturation intensities for describing the excitation are
IPsat ∫ hcbP/λPτRσP, and ILsat ∫ hcbL/λLτRσL. (5.6)
For simplicity the results will be cast in terms of dimensionless
scaled variables:
iP ∫ IP/IPsat, iL ∫ IL/ILsat, and n2 ∫ N2/NT. (5.7)
The above material parameters can be evaluated using the

well-known approximations of Fuchtbauer–Landenburg and the
McCumber reciprocity relations. With these approximations, the
temperature-dependent quantities can be computed from the
energy-level structure and effective absorption cross sections [8, 29,
49]:
1 Z (T ) È1 Ê hc ˆ ˘
=1+ 1 exp ÁË e2 - ˜¯ ˙ ,
b (l , T ) Z 2 (T ) Í
Î kT l ˚

1
=
Z1 T( ) exp Ê e2 ˆ Èexp Ê -hc ˆ - exp Ê -hc ˆ ˘ , (5.8)
bPL Z2 T( ) ÁË kT ˜¯ ÍÎ ÁË lL kT ˜¯ ÁË l kT ˜¯ ˙
P ˚
Ú s A (l , T ) exp( - hc l kT ) dl l
4
and
d lF = band
.
Ú s A (l , T ) exp( - hc l kT ) dl l
5
band
Here Z1(T) and Z2(T) are the Boltzmann thermal partition

functions of the lower and upper manifolds and e2 is the minimum
energy of the upper manifold. The quantity e2 is often referred to as
the zero-phonon energy.
With the above definitions the steady-state solution to Eq. 5.2
can be written as
iP b P + iL b L
n2 = . (5.9)
1 + iP + iL
The steady-state laser gain and pump absorption coefficients are
then simply
È iP b P b PL - 1 ˘ È 1 + iL b L b PL ˘
gL = s L NT ◊ Í ˙ and a P = s P NT ◊ Í ˙ . (5.10)
Î 1 + iP + iL ˚ Î 1 + iP + iL ˚
The above relations make clear the physical significance of the
fractional cross sections. From Eq. 5.2 we can see that b(l, T) is
simply the excitation fraction at which absorption transitions to gain
for wavelength l. From the expression for laser gain, we can see that
bPL/bP is equal to the minimum pump intensity required to produce
laser gain. Note that this is true at any laser intensity.
The expression for pump absorption is a bit more subtle. For
pump intensities below the gain threshold, pump absorption
decreases with laser intensity. Above the threshold, aP increases
with increasing laser intensity. This is because stimulated emission
competes with absorption and these two have the opposite effect on
ground-state populations. The influence of laser intensity on pump

absorption clearly scales with the value of bPL/bL.
To complete the steady-state solution, one should solve for the
powers flowing through the optical medium. The optical power
densities cycling through the laser medium for the pump and laser
fields are identified as
P = aPIP and L = gLIL. (5.11)
The steady-state power conversion directly into heat through the
laser’s effective QD is found from Eq. 5.3 to be
H=
[
NT hc lF ( lL - lP )iP iL bP bL + lL ( lF - lP )bP bPL iP - lP ( lL - lF )bL bPL iL ].
lF t R [ ]
l P l L 1 + iP + iL b P L
(5.12)
Unlike the radiative power densities, heat builds up in the lasing
medium, leading to temperature gradients and detrimental thermo-
optic distortions. For a steady-state laser, temperature gradients
must build until equilibrium is established. In a high-power laser
system, significant additional power and hardware will be required
to manage the thermal effects. It is therefore reasonable to reduce
heating, while maintaining laser performance. In the next section we
will examine laser optimization strategies using this simple model.
We illustrate the above relations using the radiative properties
of Yb:YAG. Figure 5.1 shows the energy structure and cross-sectional
spectra at room temperature. The mean fluorescence wavelength and
radiative lifetime of Yb:YAG are reported to be 1010 nm and 0.979
ms, respectively, at room temperature [5, 10]. For these calculations
the laser wavelength will be chosen at 1050 nm, which corresponds
to the lowest-energy transition from the 2F5/2 level of Yb:YAG. The
peak absorption cross section of the 1050 nm line is only 6.0E-23
cm2, leading to a saturation intensity of 70 kW/cm2.
In Fig. 5.2 we illustrate how the cross-sectional parameters
defined above vary with the choice of operating conditions in the
Yb:YAG system. It shows that the βs can vary by many orders of
magnitude for reasonable choices of operating wavelengths and
temperature. The single-wavelength parameter b(l, T) is plotted
using lines with triangles for 100 K, 200 K, and 300 K. At short
wavelengths, b(l, T) approaches unity for all temperatures, but it
falls off rapidly with wavelength and temperature. Since b(l, T) is the
minimum excitation required for gain, thresholds are significantly

reduced for low temperatures.
-19
10
Yb:YAG @ 293 K (a) 10930 (b)

2F 10624
5/2
Effective Cross Section (cm )
2
-20 10327
10
λP λL
-21
10
-22 λP λL 785
10
2F 612
7/2 565
0 cm-1
880 920 960 1000 1040 1080 Yb:YAG
Wavelength (nm)
Figure 5.1 (a) Room-temperature absorption (lP) and emission (lL) cross
sections measured in optically thin samples of Yb:YAG. (b)
Yb:YAG energy levels at room temperature.
10
0.1
0.01
0.001
0.0001
0.00001
0.9 0.92 0.94 0.96 0.98 1
lP/lL
Figure 5.2 Plot of the fractional cross sections discussed in the text
evaluated for Yb:YAG. Lines with triangles delineate βP. Lines
with circles are βPL/βP and solid lines are βPL/βL. Values are
calculated for several temperatures, assuming λL = 1050 nm:
300 K (red), 200 K (green), and 100 K (blue).
Room-temperature Yb:YAG requires a minimum excitation of 2%

at 1050 nm, but at 100 K the threshold drops by 2000-fold. Note that
the curves for all b(l, T) cross at 968 nm. This is the zero-phonon
transition of Yb:YAG; l00 = hc/e2.
Figure 5.2 also plots the parameters bPL/bP and bPL/bL. As
discussed above, bPL/bP is the pump intensity required to create an
inversion and is shown here as dashed lines with open circles. At
short pump wavelengths, bPL/bP approaches bL, the laser excitation
threshold discussed above. However, as the pump wavelength is
increased, thresholds grow monotonically and then diverge to
infinity as the pump wavelength approaches the laser wavelength.
At longer wavelengths higher pump levels are required to offset the
effects of pump saturation.
The values of the bPL/bL parameter in Fig. 5.2 are plotted with
solid lines. This parameter controls the influence of laser intensity
on pump absorption and laser efficiency. Note that bPL/bL is always
slightly greater than unity. As a result material performance is
strongly coupled to laser intensity. Nevertheless, Fig. 5.2 reveals
that bPL/bL is only weakly dependent on pump wavelength and
temperature. Only as the pump wavelength approaches that of the
laser does bPL/bL rise sharply. Larger values of bPL/bL increase pump
absorption, but as we will see in the next section, it decreases optical
efficiency.
5.3.2 Optimization of Laser Material Performance

The steady-state power densities derived in the previous section
allow us to find the optimal performance conditions of the ideal
laser medium. For this purpose, we define laser material efficiency
as the emitted laser power divided by the absorbed pump power:
hm ∫
L
=
[
lP 1 - bPL bP iP ] . (5.13)
P [
lL 1 + bPL bL iL ]
Note that the dependence of the efficiency on intensity has
shifted from that in Eq. 5.10. Now increasing either the pump or the
laser intensity increases hm.
We should be careful to distinguish the laser material efficiency,
hm, from the more commonly used laser optical efficiency. The laser
optical efficiency will be proportional to hm but will also depend
on beam propagation through an optical system. Equation 5.13

provides an upper bound to optical efficiency that a well-designed
laser system should approach. Of course, the appropriate spatial
average must be applied to the local hm solution.
To compare the magnitude of the laser internal heating, we will
define a laser material heating parameter:
Xm ∫
H
=
[lF (lL - lP ) iPiL + lL (lF - lP ) iP bPL bL - lP (lL - lF ) iL bPL bP ].
L lF lPiL [iP - bPL bP ]
(5.14)
Recognizing that the heat generation can be positive or negative,
we define Xm in terms of the absolute magnitude. Note that Xm is
defined in terms of laser power and not absorbed pump power. The
heating parameter more commonly used in the literature would be
the product of hm and Xm. The definitions used here will lead more
directly to an optimization procedure.
The above equations describe the performance of a purely
radiative cw laser. Laser power can theoretically be optimized by
either maximizing hm or minimizing Xm. This would be accomplished
through the appropriate selection of the operating intensities and
wavelengths. Note that the heating fraction depends strongly on
wavelength separations, while the laser efficiency is independent of
the fluorescence wavelength.
Consider first the approach of improving laser performance
by maximizing the laser material efficiency. Equation 5.13 reveals
that whenever the laser material is pumped far above inversion, the
material efficiency will approach lP/lL. Lower intensities reduce the
material efficiency and longer pump wavelength generally increase
it. However, we can see from Fig. 5.2 that as lP approaches lL, bPL/
bP and bPL/bL increase rapidly, raising the laser’s threshold and
decreasing efficiency. This optimization approach is illustrated in
Fig. 5.3 for the idealized Yb:YAG lasing at 1050 nm. As expected, laser
material efficiency initially rises linearly with pump wavelength and
then falls to zero as lP approaches lL. Good efficiencies require that
intensities exceed the saturation value. At the highest intensities
considered, the room-temperature material efficiency reaches 88%
when the pump wavelength is 1005 nm. Cooling is seen to improve
material efficiency somewhat at longer wavelengths. For the highest
intensity plotted, the 100 K material efficiency is predicted to reach a

maximum of 91.5% when pumped at 1028 nm. This simple analysis
suggests that for sufficiently high intensities, the material efficiency
can approach arbitrarily close to unity.
However, this analysis ignores the increased thermal loading
that can occur at very high intensities.
An alternative approach to improving laser performance would
be to minimize the heating parameter Xm. This approach will generate
the least amount of waste heat per watt of laser output power. Figure
5.4 shows Xm calculated for the room-temperature conditions that
were plotted in Fig. 5.3. In contrast to the slowly varying laser
efficiency, the heating parameter displays large variations. There is
a null in the heat generation that shifts to longer pump wavelengths
as the intensities rise. As the pump wavelength is tuned through this
point, heat generation switches over to optical cooling. For longer
pump wavelengths, Xm increases rapidly exceeding unity above the
laser threshold.
The figures above were evaluated for simplicity only at iP = iL, but
in general the pump and laser intensities will vary independently.
While the equations are symmetric in the two fields, the large
difference in bP and bL breaks this symmetry. For pump wavelengths
shorter than lF, both efficiency and heat parameters depend almost
entirely on the laser intensity. For pump wavelengths greater than
lF, both intensities determine the roll-off and asymptotic values.
Nevertheless, the basic shape of both curves remains unchanged at
all intensities. The null in the heating efficiency is referred to as the
RBL condition.
At the RBL operating point the fluorescence cooling is precisely
offset by the laser heating [12, 14]. It is easy to show that when Xm
= 0, the several simple relations must be true. Firstly, the pump and
laser intensities must satisfy a simple relationship determined by
the QDs:
iPmin lP (lL - lF ) bPL

iP = where iP min ∫
1 - iLmin iL lF ( lL - lP ) bP
(5.15)
lL (lP - lF ) bPL
and iLmin ∫ .
lF ( lL - lP ) bL
1.0
0.8
0.6
ηm
0.4
i = 16
= 8
= 4
0.2 = 2
= 1
= 0.5
0.0
0.90 0.92 0.94 0.96 0.98 1.00
λP / λL
Figure 5.3 Plot of the material efficiency for an ideal Yb:YAG laser with
λL = 1050 nm and several values of the scaled intensity. For
the purpose of this illustration, the scaled pump and laser
intensities are set equal. Solid curves are for 300 K and dashed
curves are for 100 K.
1
λF
0.1
0.01
m
Ξ
i = 16
= 8
= 4
0.001 = 2
= 1
= 0.5
0.0001
0.9 0.92 0.94 0.96 0.98 1
λP / λL
Figure 5.4 Plot of laser material heating parameter for idealized Yb:YAG.
Lasing is at 1050 nm and the temperature is 300 K. The vertical
line marks the position of the mean fluorescence wavelength.
As before, the scaled pump and laser intensities are set equal.
So the RBL pump and laser intensities are inversely related.

As the laser intensity increases, the pump intensity approaches a
minimum allowed value, iPmin. Conversely, there is also a minimum
laser intensity required to maintain the radiation balance at high
pump intensities. Since real RBL systems are likely to be limited by
the pump brightness, operation just above iPmin is likely to be the
optimal design.
It is also easy to show that at the RBL point the laser efficiency
reduces to
( lP - lF )
hRBL = . (5.16)
( lL - lF )
This simple expression reveals that as the pump wavelength ranges

between lF and lL, hRBL varies from 0 to 1.
To see how the RBL efficiency depends on the pump intensity.
Let’s assume that the laser intensity is high, so the RBL pump
intensity will be close to iPmin. Figure 5.5 illustrates how hRBL varies
with the minimum pump intensity for the example of a room-
temperature 1050 nm Yb:YAG laser. From Fig. 5.5 we see that when
iPmin = 10, the RBL efficiency should approach 80%. However, to get
hRBL > 90% would require 10 times higher pump intensities.
1 0.05
0.8 0.04
Quantum Defect
0.6 0.03
(lP – lF)/lP
ηRBL
0.4 0.02
(lL – lP)/lL
0.2 0.01
0 0
3
0.1 1 10 100 1×10
iPmin
Figure 5.5 The solid curve is the predicted RBL material efficiency as a
function of the minimum pump intensity. The dashed lines
give the laser quantum defect (lL – lP)/lL and the fluorescence
quantum defect (lP – lF)/lP. This calculation assumes the
conditions of the Yb:YAG laser discussed in the text.
The dashed curves in Fig. 5.5 illustrate how the QDs for the laser
and fluorescence emission vary with minimum pump intensity.
When the QDs are equal the heat generated by one laser photon
is on average offset by the cooling from one fluorescent photon.
At this point the RBL efficiency is 50%. As the pump wavelength is
increased, laser QD heating decreases and fewer fluorescent photons
are required to maintain balance. This requires the laser efficiency
increases to maintain the RBL condition. For example, at iPmin = 10,
a single fluorescent photon should balance the defect of three laser
photons.
Comparing Figs. 5.3 and 5.4 reveals that optimums for the two
approaches do not occur at the same pump wavelength. The RBL
points occur at longer wavelengths for the same intensity. Taking
the iP = iL = 2 case, for example, an ideal 1050 nm Yb:YAG laser
pumped at 970 nm is predicted to have hm = 0.593 and Xm = 0.102.
The same laser pumped at 1026 nm would see hm = 0.402, with
no heat generation. At the highest intensities plotted, iP = iL = 16,
the maximum possible efficiency is 0.882, while the RBL efficiency
is 0.745. Since thermal loading requires active cooling to maintain
temperature and beam quality, these systems’ overall efficiencies
are likely to be comparable.
A third approach for improving laser performance is motivated
by the special case of no lasing. When laser intensities are small, the
laser material heating efficiency simplifies to become proportional
to lP – lF. So if the laser medium is pumped at its mean fluorescent
wavelength, there will be no heat generated by the pump. This
approach might be indicated when the stored energy is not fully
extracted from the laser medium, as often occurs in bulk laser
geometries or during pulsed mode operation. For a laser operating
with the condition of lP = lF, the fractional laser heating simplifies
to the laser’s QD:
Xm = (1 – lL/lF). (5.17)
Note that this is the only pump-intensity-independent heat-
loading solution for an LQDL. Heat loading would be simply
proportional to laser power. In the example plotted in Fig. 5.3 this
operating point is labeled by a vertical dashed line. Pumping at 1010
nm and lasing at 1050 nm give a heating parameter of 3.8%. Blue-
shifting the laser wavelength could readily reduce the laser heating
parameter to below 1%, while maintaining good efficiency.
5.3.3 Nonideal Laser Materials

A more accurate model of laser power flow must include realistic
material losses. In the previous section the laser material was
assumed to be perfectly radiative, linear, and optically thin. Now
we will consider what happens when those assumptions are only
mostly correct.
First we will include the possibility of weak quenching of
the laser excitation. This could occur through electron–phonon
deactivation or through an ion–ion energy transfer mechanism
[66]. Direct quenching introduces another linear loss term into Eq.
5.2. This linear loss term is usually handled simply by replacing the
radiative lifetime, tR, with a fluorescence lifetime, tF. Energy transfer
losses are effectively handled by adding a quadratic loss term with a
characteristic quenching lifetime, tQ. For simplicity, we will assume
that the quenching process is completely nonradiative; quenched
excitations are converted completely into heat.
The second process to be added to the model is background
absorption. Background absorption is present at low levels in even
the best optical materials [13]. Optical cooling studies have shown
that these weak losses become important for low-QD pumping [36].
The background absorption, aB, is usually assumed to be due to trace
impurities, so the excitation model is not directly affected by this
process. Here we will also assume that the background absorption
is sufficiently broad to affect both pump and laser. Define fractional
absorptions as daP = aB(lP)/sPNT and daL = aB(lL)/sLNT. As we will
see, even small background absorptions can significantly reduce
efficiency and increase heating.
The last process needed for a more realistic laser model is
reabsorption of fluorescence inside the laser medium. Often
referred to as radiation trapping, this process will occur whenever
absorption or reflection prevents some of the spontaneous emission
from escaping the laser medium. We will assume that reabsorption
is due solely to the active ions. So while trapping can enhance other
nonradiative losses by increasing excitation densities, here it is
considered a purely radiative phenomenon.
The addition of radiative trapping significantly complicates the
modeling of the laser materials. It introduces nonlocal source terms
for the fluorescent field that require intense numerical calculation
for solution. For this analysis, we will invoke an approximate solution

for the trapped fluorescence field [11]. The derivation and general
solutions of the revised rate equation are presented in the appendix
to this chapter. The approximation assumes that the laser medium
traps light uniformly throughout and that the amount of trapping
can be described simply by a cavity photon lifetime. This allows the
fluorescent field to be removed from the rate equation. The rate
equation then simplifies to a quadratic in terms of the excitation
density and modified values for fluorescence wavelength, l¢F, and
lifetimes, t ¢F, and t ¢Q. These are the effective values that would be
measured external to the optically thick laser medium.
It should be noted here that it is not usually possible to measure
the radiative lifetime, tR, directly. In the literature it is generally
calculated from the absorption or emission spectra. When quenching
is weak a more accurate measurement of tR is possible by measuring
nonradiative losses directly. To obtain the radiative lifetime of
Yb:YAG quoted above, we used a low-power tunable laser to perform
photothermal deflection on bulk samples [5]. This technique allows
an accurate measurement of the pump wavelength at which optical
heating crosses over to optical cooling, lC. The radiative lifetime
can then be computed by correcting the measured fluorescence
lifetime for nonradiative losses using the relation tR = tFlC/lF. The
best fluorescence lifetime measurements are acquired using thin
powdered samples immersed in an index-matching medium to avoid
fluorescence trapping.
Incorporating these loss processes into the revised rate equation
allows us to solve for the power densities in a realistic laser material.
We will distinguish this solution from that of the ideal material by
adding a prime. The revised rate equation now becomes
dN2¢ N2¢ N2¢
2
I P l Ps P N¢ IL l L s L N¢
= ◊ È NT - 2 ˘+ ◊ È NT - 2 ˘- - .
dt hc ÍÎ bP ˙˚ hc ÍÎ b L ˙˚ t F¢ N Ybt Q¢
(5.18)
While the general solution is presented in the appendix, here
we continue with the approximation of weak losses. With that
assumption, the steady-state solution can be written in terms of the
ideal solution, n2, given by Eq. 5.9:
È (1 - t R t F¢ ) - n2 t R t Q¢ ˘
n2¢ = n2 ◊ (1 + d n ¢ ) ª n2 ◊ Í1 + ˙ . (5.19)
Î (iP + iL + t R t F¢ ) ˚
This expression makes it is clear that steady-state excitation

densities depend principally on the change in the fluorescence
lifetime. A decrease in the lifetime from linear quenching will
generate a proportional decrease in the excitation density.
Conversely, an increase in t ¢F due to radiative trapping will increase
excitation densities. For many materials the effect of trapping is
more pronounced, leading to longer fluorescence lifetimes and
higher excitation densities in optically thick samples. Note that since
n2 << 1, the quadratic quenching term will have little effect on the
density unless t ¢Q ≈ tR.
Using the same fractional definition as before, the laser material
efficiency is modified by the revised excitation fraction and by the
additional absorption:
¢
Lm È n2¢ - bL - bLda L ˘
l P iL
¢ ∫
hm = Í ˙ . (5.20)
Pm¢lL iP Î bP - n2¢ + bPda P ˚
In the approximation of weak losses, it becomes
È ¢ - bPda P
nmd nm ˘ ¢ - bLda L
nmd nm
¢ ª hm ◊ Í1 +
hm + ˙ . (5.21)
Î bP - nm ˚nm - b L
Expressed in this way, it is clear that laser material efficiency

increases with excitation density and decreases with background
absorption. The contribution from background absorption is
reduced by b. The impact of these losses on efficiency increases
substantially when the material is operated near the laser threshold
or pump saturation.
To illustrate the impact of including material losses we again
use Yb:YAG as the example. This time we assume small realistic
corrections to the material parameters. In the slab laser experiments
discussed in the next section, it was found that trapping increased
the fluorescence lifetime to 1.25 ms and the mean fluorescence
wavelength shifted to 1017 nm. We will also assumed weak quadratic
quenching by setting t ¢Q to 20 ms and moderate background
absorption with daL = daP = 10–3. These values are consistent with
values observed in samples of high-purity laser materials with low
Yb-doping densities [13, 36].
Figure 5.6 shows the result of the evaluation of Eq. 5.21 for the
same conditions used in Fig. 5.3. With the more realistic material
parameters given above, efficiencies are higher than for the ideal
case. This is due to the increased fluorescence lifetime produced

by the radiative trapping. The impact is larger at lower intensities
where efficiencies increase by roughly 10%. At the highest intensity,
room-temperature optimal efficiency rises only slightly to 0.895 at
1008 nm. For operation at 100 K, the optimal efficiency is 0.927 at
1029 nm. Still, the efficiency curves are relatively flat, allowing wide
latitude in the selection of the pump wavelength.
1.0
0.8
0.6
η´m
0.4
i = 16
= 8
= 4
0.2 = 2
= 1
= 0.5
0.0
0.90 0.92 0.94 0.96 0.98 1.00
λP / λL
Figure 5.6 Plot of the predicted laser material efficiency for the same
conditions used in Fig. 5.3. Material parameters have been
modified, as described in the text.
The form of Eq. 5.21 also suggests a strategy to maintain

efficiency in the face of higher background absorption. To a limited
extent, background absorptive losses will be offset by radiative
trapping. Fluorescent trapping significantly increases storage
lifetime, excitation density, and laser material efficiency.
In a similar way we can revise the laser heating parameter
that incorporates the weak loss processes in the laser material.
Heat generation will be modified differently by each of the loss
processes. Since radiative trapping is assumed to be purely radiative
its contribution to laser heating can be computed directly from
the steady-state solution in Eq. 5.14 using the measured effective

material properties. The mean fluorescence wavelength, l¢F, should
be measured using the actual laser medium to fully account for the
geometrical effects of trapping. The measured fluorescence lifetime
should be corrected to remove quenching using the relation, t ¢R =
t ¢FlC/lF. Substituting l¢F and t ¢R into Eq. 5.14 will yield the QD heat
generation corrected for the fluorescent reabsorption, H¢F.
In addition to the QD heating, we identify specific contributions
for the heat generated by quenching, dH¢Q, and by background
absorption, dH¢A:
hcNT n2¢ Ê 1 1 n2¢ ˆ
d HQ¢ = Át - + ˜
lP Ë F tR t Q¢ ¯
(5.22)
hcNT Ê iP b P iL b L ˆ
and d H A¢ = ÁË l da P + l da L ˜¯ .
tR P L
The total heat generation in the realistic laser material now
becomes H¢ = H¢F + dH¢Q + dH¢A. Defining the material heating
parameter as before leads to:
¢ ∫
Xm
H¢ t
= F
[lF¢ (lL - lP ) iPiL bP bPL + lL ( lF¢ - lP )iP bP bL - lP ( lL - lF¢ )iL ]
L¢ t F¢ [
lF¢ lPiL iP bP bPL - 1 ]
l Ê tR n t ˆ È i b i b + 1 ˘ È iL bLda L + iP bPda P lL lP ˘
+ L ÁË t-1+ 2 R ˜ Í P P L L ˙+Í ˙.
lP F t Q¢ ¯ Î i P bP bPL - 1 ˚ Î n2 - bL ˚
(5.23)
Here the first term on the right accounts for radiative trapping.
The second term accounts for quenching and the third accounts for
background absorption. Here the approximation, n¢2 ≈ n2, found to
be valid in Eq. 5.21 has been used. Figure 5.7 plots X¢m for the Yb:YAG
example for the same conditions used in Fig. 5.6. Comparing this plot
with that of the idealized Yb:YAG in Fig. 5.4 reveals the impact of
adding realistic material losses. Linear and quadratic quenching red-
shift the crossover wavelength by 5 nm and background absorption
adds another 1 nm shift. When radiative trapping is added to the
model t ¢F = 1.25 ms, thermal loading decreases and the crossover
wavelength red shift increases to 15 nm. This narrows and disrupts
the RBL resonance at lower intensities. Background absorption
heating is stronger at higher intensities. For the highest intensities,
a fractional background absorption of daL = daP = 10–3 disrupts the
RBL resonance. Increasing the background absorption fraction to 4

× 10–3 was found to overwhelm fluorescence cooling, disrupting the
RBL resonances at all wavelengths.
Comparing Figs. 5.4 and 5.7 clarifies the impact of weak losses
on optimal material performance. The large red shifts in the RBL
resonances significantly reduce material efficiency at minimum X¢m.
The best efficiency with no heat generation is 0.40 when the pump
is 1040 nm with intensities of i = 4. Operation away from the exact
RBL resonance allows material efficiencies to recover. If we restrict
X¢m ≤ 0.01, a best efficiency 0.62 occurs for i = 8 with the pump at
1040 nm.
Figure 5.7 Plot of the realistic room-temperature laser material heating

parameter for the same conditions used in Fig. 5.4. The vertical
dashed line marks the position of the effective crossover from
fluorescent heating to fluorescence cooling, l¢C.
5.4 Low–Quantum Defect Laser Experiments
5.4.1 Introduction to Experiment Review

In the previous section we considered a generic laser material that
consisted of two widely separated sets of discreet energy levels.
Low–Quantum Defect Laser Experiments 169
The discreet energy levels were treated as existing in a thermal

equilibrium with their surroundings. The upper and lower manifolds
were assumed to be separated sufficiently such that thermal
transitions between them were unlikely. It is this condition that
allows for the design of laser systems with low QDs. As discussed
earlier, these conditions are satisfied in the trivalent ytterbium
system. Ytterbium ions have a single pair of 4f states that are roughly
3 kT wide and are separated by approximately 50 kT at room
temperature. This energy splitting and the absence of other energy
levels make the Yb3+ system ideal for LQDL and optical cooling.
In this section we will review a few of the many laser experiments
conducted with LQDL systems. The first section will focus on the 1
mm ytterbium system, particularly results approaching radiation
balance. Alternative active ions are considered separately. For these
ions the presence of additional excited states complicates the simple
model developed earlier. Nevertheless many of these systems are
also promising candidates for optically cooled lasers.
5.4.2 Ytterbium Laser Experiments

Since its first demonstration, ytterbium has been lased in al-
most every host material [44]. As mentioned above, the ytterbium
1 mm transition is spectrally broader than most other rare earth tran-
sitions. This makes Yb3+ ideal for LQDL designs but often leads to
lower absorption and emission cross sections. As shown in the previ-
ous section, weak absorption and gain tighten the requirements on
material losses and increase the need for high-intensity pump sourc-
es. Successful LQDL systems are therefore based on the highest–opti-
cal quality materials. After decades of process refinement, YAG and
fused silica have the highest optical quality and lowest losses. As a
result the best LQDLs have so far been based on these materials.
Most ytterbium lasers have utilized GaAs laser diodes as the
pump source. Indeed, it was the invention of bright laser diode
pumps that made Yb3+ a practical laser system. Nevertheless the
laser diode’s limited brightness and restricted operation wavelength
have hampered LQDL design. As a result, most of the experiments
have used other pump sources to generate the required intensities
and pump wavelength. In this section we will review a few of the
more successful LQDL ytterbium lasers.
In 1999 the Naval Research Laboratory began a series of

ytterbium LQDL experiments. Early laser designs explored direct-
diode pumping of ytterbium-doped KGd(WO4)2 crystals [11, 13].
This biaxial crystal is attractive for LQDLs because of its somewhat
blue-shifted transition and higher cross sections. In 2000 high-
power laser diodes with wavelengths approaching 1 µm were just
becoming available, so for these experiments we chose lP = 995 nm.
Several slab and disk laser geometries were tested using the high-
cross-section m axis transitions for pump absorption and the p axis
laser transition at 1045 nm. (The principal optical directions are
labeled p, m, and g in order of increasing refractive index.) Figure 5.8
shows an edge-pumped 2 × 8 mm Yb:KGW laser disk with half of its
90 radially focused laser diode bars. These Yb:KGW lasers generated
up to 490 W at moderate duty factors using only passive cooling.
This was the first LQDL system with heat loading below the laser
QD. Room-temperature laser saturation intensity for Yb:KGW is only
2.4 kW/cm2 and a high-Q cavity helped maintain 10 ≤ IL/ILsat ≤ 50.
Background absorption at 1 µm in KGW was measured to be roughly
8 × 10–4 cm–1 or daL = 4 × 10–4. With measured values of X as low
as 0.08, these experiments demonstrated the promise of LQDLs for
average power scaling.
Figure 5.8 Edge-pumped Yb:KGW disk laser with a laser quantum defect
of 4.8%.
While direct-diode pumping is attractive from a system

perspective, it led to several problems in these prototypes. First
high-power diodes were not available at the longer wavelengths

needed to generate optical cooling. Further diode coupling was
limited to around 55%. For optically thin Yb:KGW, lF = 997 nm and
the radiative lifetime is 0.275 ms. However, for the 2 × 8 mm laser
disk with 2.1E20 Yb density, fluorescent trapping increased these
values to l¢F = 1011 nm and 0.75 ms. Also, the on-axis intensity of the
diode pump array limited to 15 kW/cm2, while the pump saturation
intensity of Yb:KGW at 995 nm is IPsat = 14.6 kW/cm2. So the scaled
pump intensity was never much greater than unity. These factors
restricted laser optical efficiency that was not better than 17%.
To explore the radiation balance condition, experiments shifted
to longitude pumping of Yb:YAG. Ytterbium fiber lasers were selected
as pump sources in order to increase brightness and red-shift the
pump well beyond the fluorescence line center at 1010 nm. An array
of 45 ytterbium-doped double-clad fiber lasers was developed for
these experiments. Each fiber generated a near-diffraction-limited
100 W beam at 1030 ± 0.1 nm. The combined output of this fiber
array produced pump intensities as high as 140 kW/cm2, or roughly
10 times higher than the earlier laser diode array. At 1030 nm the
pump saturation intensity of room-temperature Yb:YAG is IPsat = 9.05
kW/cm2. Therefore this fiber array was capable of generating pump
intensities iP ≤ 15. Figure 5.9 is the steady-state gain for Yb:YAG
computed from the spectral data in Fig. 5.1 and Eq. 5.10. This figure
assumes iL = iP, room temperature, and an Yb density of 4.1E20
ions/cm3. Dashed curves are identical data plotted at the 10 times’
scale shown on the right axis. Pump saturation reduces the 1030 nm
absorption by eightfold, even in the presence of strong stimulated
emission. Note that with increasing intensity, gain at 1049 nm
approaches, but can never exceed, sLNYbbPL/2bP, or 0.0176 cm–1.
Strong gain clamping is a common characteristic of LQDL systems.
In the first series of experiments the fiber array was used to
end-pump a single Yb:YAG rod [10]. The rods used were 3 mm in
diameter and roughly 10 cm long. The ends were AR coated and
the barrel of the rod was polished. The Yb:YAG laser rod was placed
inside a stable optical resonator, as shown in Fig. 5.10. The laser rod
was placed on a thermally insulated fixture and a thermal camera
was used to measure the surface temperature of the Yb:YAG crystal.
0.02
0
i = 16
Gain Coefficient (cm–1)
= 8.0
= 4.0
= 2.0
= 1.0
0.01
− 0.1 = 0.5
− 0.2 0
1020 1030 1040 1050 1060 1070 1080
Laser Wavelength (nm)

Figure 5.9 Gain coefficient for room-temperature 3% Yb:YAG pumped
at 1030 nm. For simplicity the pump and laser intensities are
set equal: i = iL = iP. Solid curves match the left vertical scale.
Dashed curves display the same data at the tenfold scale shown
on the right axis.
Figure 5.10 Optical configuration of the first radiation-balanced laser. (a)

Stable Yb:YAG resonator. (b) Ytterbium fiber laser pump array
propagation shown without the target laser rod.
The pump beams were injected into the resonator off-axis using
a flat HR mirror with a 3 mm on-axis hole. The broadband HR back-
resonator mirror reflected the pump beams back to the center of
the rod. The beams crossed near the center of the rod, producing a
roughly 2 mm diameter mode volume. The concave output mirror
was coated for partial reflection at 1050 nm and low reflectivity at

1030 nm. An 80% reflector was found to generate maximum power.
The cavity length was also adjusted for maximum power. A Brewster
plate was added to ensure linearly polarized operation.
The resonator above was operated under a variety of conditions.
Laser rods with target ytterbium densities of 1%, 2%, 3%, and 5%
were tested. Pump power, output coupling, mirror curvature, and
cavity length were adjusted while monitoring output power, thermal
loading, and beam quality. Typical steady-state laser performance is
illustrated in Fig. 5.11. Somewhat higher efficiencies were obtained
with the 5% doped crystal, owing to its higher pump absorption.
Lower thermal loading and better beam quality were found with the
2% doped crystal. This resonator generated average powers up to
200 W of linearly polarized output with a beam quality of M2 = 1.2,
as illustrated in Fig. 5.12. During laser operation the YAG rod was
thermally insulated and monitored with the thermal camera.
100
90
Laser Power (W)
Slope = 60%
Slope = 41%
80
70
60
50
280 300 320 340 360 380 400 420
Incident Pump Power (W)
Figure 5.11 Comparison of the laser performance of the Yb:YAG resonator
for two doping densities: 3.1E20 ions/cc (dots) and 8.1E20
ion/cc (squares).
For the above laser configuration the minimum heat load condition
was found to require an incident pump intensity of iP = 2.1 and an
on-axis intercavity laser intensity of iL = 1.0. This is a close to the
predicted value of the nonideal model developed in Section 5.3.3.
This is consistent with the strong radiation trapping observed in
these experiments. Fluorescence lifetimes from these laser rods were
found to be 2.0 ms, twice as long as the measured value τR. Near the
RBL resonance, the laser power was 80 W and optical efficiency was
26%. This is consistent with the nonideal LQDL model that predicts
the average material efficiency for the rod should be 27%.
Figure 5.13 shows a typical thermal image of the laser rod near
the RBL condition. For optimized conditions heat loading in the
laser rod was too low to measure with the thermal camera. This
corresponds to a material heat parameter X´ < 10–4. This was the
first truly RBL.
250
200
Power (W)
150
100
50
0
0 10 20 30 40
Time (s)
Figure 5.12 Power and near-field beam profile of the 3% Yb:YAG resonator.
Figure 5.13 A typical thermal image of the 3 × 120 mm Yb:YAG rod inside
the radiation-balanced laser. This image was captured after
10 seconds of lasing while being pumped with 372 W at 1030
nm. The false color scale corresponds to ±5°C, as shown on the
right. Fluorescence from the laser rod heats the mirror mount
on the left, but the thermally insulated laser rod remains very
close to the ambient temperature.
A second series of experiments was conducted to improve

performance of the Yb:YAG laser system [52]. In these experiments
we replaced the laser rod with a composite slab composed of Yb:YAG
and sapphire. A single 2 × 110 mm 3% Yb:YAG slab was fusion-
bonded to a sapphire plate along its length, as shown in Fig. 5.14.
The height of the YAG slab was tapered along its length from 3 mm
to 2 mm and an undoped 10 mm end-cap was bonded to the larger
end.
Figure 5.14 Radiation-balanced laser gain element constructed from a

tapered 110 mm 3% Yb:YAG slab fusion bonded to a 2 × 20 mm
sapphire plate. The sapphire plate provides a refractive index
mismatch of 0.07. This allows for guiding of the 1030 nm pump
lasers, while significantly reducing fluorescent trapping. The
gold-plated mount and fused silica pins were designed to avoid
radiative heating of the fixture.
The composite slab provided several important improvements for

the RBL. It provided a rugged but transparent fixture. The polished
edges provided guiding of the fiber pump beams down the length of
the YAG. This resulted in more uniform pumping and higher overall
pump absorption. The small difference in refractive index between
YAG and sapphire produced less trapping of the Yb fluorescence.
The sapphire bond reduced the fluorescent trapped lifetime in the
slab to 1.25 ms, from the 2 ms measured in the freestanding rods.

The tapper helped suppress parasitic modes and smoothed pump
intensity.
The slab was end-pumped through the AR-coated end-cap. Pump
powers as high as 4.8 kW were incident on the 2 × 3 mm facet, leading
to scaled pump intensities of iP ≤ 8.5. As with the rod experiments, an
80% and high-reflection mirror created a stable resonator around
the slab. Laser powers at 1050 nm up to 1.6 kW were generated
under pump-limited conditions, leading to intercavity intensities of
iL ≤ 5.1.
Minimum heating was observed at a laser power of 600 W,
where the measured heating parameter dropped to below 0.001.
That corresponds to 1 W of heat for every kilowatt of output laser
power. This RBL operating point corresponded to an incident
pump intensity of iP = 4.7 and an intercavity laser intensity of iL =
1.9. For the conditions of this experiment the nonideal LQDL model
predicts the RBL condition will occur at iP = 3. The measured optical
efficiency of the laser was 33% and the model predicts a spatially
average efficiency of 31%.
These experiments illustrate the importance of increasing
pump intensity and reducing fluorescent trapping. Index matching
significantly reduced trapped lifetimes. This shifted the RBL
resonance for 1030 nm pumping to higher intensities that increased
the RBL power by 7.5 times. They also illustrate the need to scale
laser mode volume to maintain the RBL intensities at the highest
pump power.
In 2007, Matsubara et al. demonstrated a record-high optical
efficiency of 81% in a cw ytterbium laser. Operating at 1048 nm, the
lasing medium was a thin disk of 20% Yb:YAG pumped by a high-
intensity 940 nm pump. At that pump wavelength the saturation
intensity in room-temperature Yb:YAG is 26 kW∕cm2. For this
demo a titanium-doped sapphire laser was used to generate pump
intensities in the disk as high as 200 kW/cm2. With values of lP/lL
= 0.90 and iP = 7.1, this demo efficiency matches well the h´m = 82%
calculation shown in Fig. 5.6. The QD heat of 10% limited average
power scaling. A thermal lens of only 5 mm was reported at the 800
mW power level.
In 2008 the group at IPG Photonics reported both high efficiency
and high power using an LQDL approach in ytterbium-doped silica
fibers [31]. They demonstrated 75% optical efficiency in a single-

mode, double-clad fiber amplifier that generated 6 kW of average
power. This system lased at 1070 nm and was pumped by several
ytterbium fibers operating at 1018 nm. Ytterbium cross sections
in silica are low, leading to IPsat and ILsat of 26 and 54 kW/cm2,
respectively [51]. Nevertheless, the reported maximum pump and
laser intensities were 20 and 2000 MW/cm2, respectively. The
lossless material model of the previous section predicts that hm =
95% in this high-intensity amplifier. This system was operated very
near the lP = lF condition, so the heat loading should be close to
the laser QD, 4.8%, when the laser efficiency approaches hm. In that
case, thermal loading would be only 20 W/m, although heat loads
were not reported. The overall electrical efficiency of this system
was reported to be 25%. Further power scaling was limited by fiber
nonlinearity, not heat generation.
5.4.3 Alternative Low–Quantum Defect Laser Materials

The conditions required for LQDL operation are also satisfied by
several other rare earth ions: erbium, thulium, and holmium. The
first excited states of these ions also decay principally radiatively in
most host materials. However, the presence of higher-lying energy
levels and lower energy transitions for these ions does give rise to
several weak loss processes. Optical cooling has been demonstrated
in Er3+ and Tm3+ but not as yet in holmium [9, 22 , 23, 28, 38, 57].
Efficient resonant-pumped lasers have been demonstrated in all
three of these ions.
Resonant pumping of erbium lasers at 1.5 mm has received
considerable attention due to the laser’s eye-safe nature. Shen et
al. demonstrated a 60 W Er:YAG laser end-pumped with an erbium
silica fiber laser at 1.532 mm [60]. The 1.645 mm laser generated an
optical slope efficiency of 81% with a QD was 7%. Although optical
cooling is not possible with this QD, heat loading was reported to be
low. Pump intensities in this demo were quite high, 200 kW/cm2.
Higher-power Er:YAG lasers pumped with 1.5 mm diodes have been
demonstrated, but weak pump absorption has so far limited their
optical efficiencies [32, 45].
Low-QD thulium 2.0 mm lasers have received less attention than
erbium. This is principally due to the lack of pump diodes at 1.9 mm
and the success of the 2-for-1 pumping scheme using 0.8 mm diodes.
Nevertheless, a resonant-pumped thulium laser system has recently
been reported with slope efficiencies of 90% [24]. This system used
0.8 mm diodes to pump a thulium fiber laser operating tuned to
1.908 mm. This fiber was then used to pump another thulium fiber
lasing at 2.005 mm, leading to a laser QD of 4.8%. As with the erbium
demo, fluorescent cooling of thulium was not possible with this
pump wavelength.
One of the best-studied midwave LQDL systems is based on
the 2.1 mm holmium transition. First demonstrated in 1992, this
laser is usually excited with a diode-pumped 2.0 mm thulium
laser, as discussed above [16, 18, 62]. This laser scheme provides
coherent beam combination of multiple thulium lasers and avoids
strong upconversion that occurs in concentrated Tm and Tm/
Ho laser materials [15]. The holmium LQDL scheme has received
considerable attention due to relatively eye-safe wavelength of 2.1
mm and the superior transmission through the atmosphere.
In 2015 Loiko et al. demonstrated record-high efficiency in a
holmium-doped KLu(WO4)2 microchip laser [47]. They achieved
84% slope efficiency at 2.105 mm pumping with a thulium laser at
1.945 mm at only 4.4 kW/cm2. Unfortunately optical efficiency was
low for this demo due to poor pump absorption and operation near
threshold.
A low-power experiment demonstrated 72% slope efficiency
from a 2080 nm holmium-doped ZBLAN fiber. The thulium fiber
pump source was operated at 2051 nm, leading to a very low laser
QD of 1.4% [35]. This LQDL scheme has the potential for optical
cooling since lF = 2010 nm in Ho:ZBLAN.
Other recent laser demonstrations have explored holmium
doped into silica fibers. Average powers up to 400 W have been
generated but with lower optical efficiencies [54, 37]. Current work
on this material seeks to reduce the hydroxyl-induced background
absorption and lower the multiphonon quenching in holmium-
doped fibers [30, 42].
5.5 Summary
LQDLs are important because they have the potential to operate
with both high efficiency and high power. Optical pumping near the
Summary 179
laser wavelength decreases the adverse effects of the laser QD. It

also increases the effects of absorption and gain saturation. When
the pump wavelength exceeds the mean fluorescence wavelength,
optical cooling becomes an important consideration.
In the first part of this chapter a steady-state model was devel-
oped that predicts LQDL material performance for an ideal laser
medium. However, as the laser QD is decreased further weak loss
mechanisms begin to play an important role. Trapping of the fluores-
cence inside the laser medium impacts both cooling and excitation
lifetime. At high intensities, extrinsic background absorption can
become a major source of heating. Extensions to the simple model
were developed to properly account for weak loss mechanisms.
The steady-state laser model was used to compare optimization
schemes. Optical efficiency was found to depend strongly on proper
selection of pump wavelength. Higher intensities were shown to
improve efficiency. Radiative trapping was found to significantly
improve optical efficiency. Optical efficiencies as high as 90% at
room temperature appear to be achievable with laser QDs of 4%.
Even higher efficiencies are possible at cryogenic temperatures
using a 2% QD. Nevertheless optimizing efficiency alone will lead to
significant thermal loading in high-power laser systems.
Alternative optimization strategies that minimize thermal
loading were also explored. Radiation-balanced lasing was shown to
virtually eliminate heat generation at the cost of somewhat reduced
optical efficiency. RBL efficiencies of 80% appear to be achievable
at room temperature with high-purity materials and careful control
of the pump intensity. The potential for higher average power,
simplified thermal management, and photovoltaic fluorescent
recycling make the RBL an attractive power-scaling approach.
A sample of the LQDL literature was reviewed and performance
was compared with steady-state theory. Several ytterbium systems
have demonstrated optical efficiencies over 75%. Optically cooled
ytterbium lasers have demonstrated heat loads below 0.01% of the
output laser power. Record performance of the low–quantum defect
ytterbium lasers clearly demonstrates the efficacy of this approach
to power scaling. The steady-state model develop here was also
found to be in good agreement with the optimum laser performance.
Low-QD experiments have been conducted in the longer-wavelength
erbium, thulium, and holmium laser systems. These systems have
also demonstrated high efficiency but are not yet capable of scaling
to high powers.
5.6 Appendix
Radiation trapping significantly impacts the performance of LQDLs.
Here we will develop an approximate solution for the excitation
density in an optically thick laser medium [10]. Using the notation
developed in Section 5.3, the rate equation for the local excitation
density will now include a term for quadratic quenching and
reabsorption of the local fluorescence:
dN2 IPs P Ê N ˆ Is Ê N ˆ N
= Á NT - 2 ˜ + L L Á NT - 2 ˜ - 2
dt hn P Ë b P ¯ hn L Ë bL ¯ t F
(5.24)
N2 Ê I (l ) ◊ s A (l ) ˆ Ê N ˆ
- 2 + Á F ˜ Á NT - 2 ˜ .
NTt Q Ë hn F ¯Ë b (l ) ¯
Here IF(l) is the local fluorescence spectral irradiance at
wavelength l. Unlike the pump and laser fields, IF(l) will be
spectrally broad with complex directionality. The last term in the
equation is the rate of fluorescence reabsorption integrated over the
entire emission spectrum as well as all directions and polarizations.
For precise evaluation of N2, the local fluorescence intensity
should be the integrated contributions from the entire laser crystal
volume. Computing IF(l) from first principles requires a detailed
knowledge of the excitation state of the entire lasing medium, as well
as its surface reflectivity. This approach would require significantly
more computation time to build up a self-consistent steady-state
spatial solution for all three fields.
To simplify the problem, we can approximate the laser crystal as
an optical cavity in which tC is the average lifetime of the available
cavity modes. This lifetime might be computed from numerical
ray traces or treated as an empirical parameter derived from
fluorescence decay data. Fluorescence reabsorption reduces the
photon lifetime within the crystal cavity according to
1 1 c ◊ a(l )
= + . (5.25)
tP tC n
Here c is the speed of light, n is the index of refraction, and a(l)
is the absorption coefficient. For simplicity we will assume that tC
Summary 181
is independent of wavelength and that the photon lifetime can be

approximated using unsaturated absorption.
We can now approximate the fluorescent spectral energy density
inside the crystal cavity at wavelength l as the product of the local
emitted power density, the probability of spontaneous emission, and
the photon lifetime:
Ê ˆ
Ê hc N2 ˆ Á g(l ) ˜
U F (l ) = Á ◊ ◊ t P . (5.26)
Ë lt R ˜¯ Á g(l )dl ˜
ËÚ ¯
Here the intrinsic fluorescence line shape function, g(l), gives
the rate of spontaneous emission of photons with the wavelength
between l and l + dl. The line shape function is closely related to
the emission cross section,
g(l) = 8πcσE(l)/l4. (5.27)
The integral in Eq. 5.26 normalizes g(l) over the entire
fluorescence band.
The approximation expressed in Eq. 5.26 is that surface reflec-
tions, scattering, and reabsorption effectively smooth the fluores-
cent energy inside the crystal. The crystal cavity approximation
should be valid in a simply shaped laser medium that is roughly uni-
formly pumped and where amplified spontaneous emission can be
ignored.
The fluorescent spectral intensity inside the crystal can be
computed using the relation IF = cUF/n and substituted back into
Eq. 5.24, eliminating the nonlocal dependence on the fluorescence
field. The rate equation for the radiation-trapped system is rewritten
as
dN2 IP s P Ê N ˆ I s Ê N ˆ N N2
= ◊ Á NT - 2 ˜ + L L ◊ Á NT - 2 ˜ - 2 - 2 (5.28)
dt hn P Ë bP ¯ hn L Ë bL ¯ t F¢ NTt Q¢
which is identical to Eq. 5.19. The definitions of the primed lifetimes
fall out of the spectral integral
1 1
∫ ◊
Ú
g¢(l ) dl
and

t F¢ t F
Ú
g(l ) dl
(5.29)
a(l )
1 1 c tC Ú b( l )
g¢(l ) dl
∫ + ◊ .
t Q¢ t Q nt F
Ú g(l ) dl
We recognize that t´F is the effective fluorescence lifetime of the

radiatively trapped system. As expected, t´F increases from tR as the
fluorescence reabsorption increases. Note that t´F is independent of
the excited-state density, so it can be measured using fluorescence
from the entire crystal. Further, t´F depends on the geometry of
the crystal cavity only through the parameter tC. This means that
a measurement of the effective fluorescence lifetime uniquely
determines the crystal cavity lifetime. The effective lifetime tQ́
combines the nonexponential decay from radiation trapping and
quadratic quenching. Here increased fluorescence reabsorption
decreases tQ́, leading to faster excitation decay. If the value of tC
is known, the integral in Eq. 5.29 can be computed and the actual
nonradiative lifetime can be extracted from the measurements of t´Q.
With these definitions the effective fluorescence line shape
function becomes
g( l )
g ¢( l ) = . (5.30)
[1 + a ( l )c t C n]
The effective line shape gives the rate of spontaneous emission
photons with wavelength l that will escape from the laser medium.
These expressions make it clear that fluorescence trapping
impacts all aspects of laser performance. Trapping increases the
energy storage lifetime and excitation densities. This leads to higher
gain and more stimulated emission rates. Stimulated emission
competes with fluorescence, so optical cooling is decreased.
Increasing the quadratic quenching rate also decreases cooling.
Trapping diminishes and red-shifts the escaping fluorescence that
also reduce cooling efficiency.
In the earlier section we solved for a steady-state solution. For
completeness we can also solve for an analytic time-dependent
solution. When the pump and laser intensities are turned on and
constant after t = 0,
1 - exp( - A ◊ t )
N2 (t ) = NSS ◊ (5.31)
1 + B ◊ exp( - A ◊ t )
with the following definitions:
A ∫ C 2 + 4 (iPbP + iL bL ) t Q¢ t R , C ∫ (iP + iL ) t R + 1 t F¢ ,
(5.32)
B ∫ ( A - C ) A + C and NSS = NT ◊ ( A - C ) ◊ t Q¢ 2.
References 183
This local solution can now be readily extended to the entire

laser medium using a numerical grid evaluation.
It is also useful to solve for the special case of IP = IL = 0. For an
initial excited state density, N0, the solution to Eq. 5.28 is the simple
decay function -1
ÈÊ N0t F¢ ˆ Ê ˆ N0 t F¢ ˘
N2 (t ) = N0 ◊ ÍÁ 1 + t ˙ . (5.33)
˜ exp ÁË t ¢ ˜¯ -
ÍÎË NTt Q¢ ¯ NTt Q¢ ˙
˚
F
This function can be fit to measured fluorescence waveforms

from the laser medium in order to determine the effective lifetimes
of a specific system. The initial excited-state density needed for the
fit can be calculated from the laser pump conditions.
References
1. Abate, J. A., Lund, L., Brown, D., Jacobs, S., Refermat, S., Kelly, J., Gavin, M.,
Waldbillig, J., Lewis, O. (1981). Active mirror: a large-aperture medium
repetition rate Nd:glass amplifier, Appl. Opt., 20(2), pp. 351–361.
2. Albrecht, G., Sutton, S., George, V., Sooy, W., Krupke, W. (1998). Solid
state heat capacity disk laser, Laser Part. Beams, 16, pp. 605–625.
3. Albrecht, G. F., Eggleston, J. M., Ewing, J. J. (1986). Design and
characterization of a high average power slab YAG laser, IEEE J.
Quantum Electron., QE-22, pp. 2099–2106.
4. Beach, R. J. (1996). CW theory of quasi-three level end-pumped laser
oscillators, Opt. Commun., 123, pp. 385–393.
5. Biswal, S., O’Connor, S. P., Bowman, S. R. (2006). Non-radiative losses in
Yb:KGd(WO4)2 and Yb:Y3Al5O12, Appl. Phys. Lett., 89(9), pp. 0919111–
0919113.
6. Bourdet, G. L. (2000). Theoretical investigation of quasi-three-level
longitudinally pumped continuous wave lasers, Appl. Opt., 39(6), pp.
966–971.
7. Bowman, S. R. (2015). Optimizing average power in low quantum
defect lasers, Appl. Opt., 54(31), pp. F78–F84.
8. Bowman, S., Ganem, R. J., Brown, C. G. (2014). Optical cooling in multi-
level systems, in Proceedings of the Laser Refrigeration of Solids VII,
SPIE 90000G, doi:10.1117/12.2040073.
9. Bowman, S. R., O’Connor, S., Condon, N. J., Friebele, E. J., Kim, W., Shaw,
B., Quimby, R. S. (2013). Non-radiative decay of holmium-doped laser
materials, in Proceedings of the Laser Refrigeration of Solids VI, SPIE

863803, doi:10.1117/12.2007511.
10. Bowman, S. R., O’Connor, S. P., Biswal, S., Condon, N. J., Rosenberg,
A. (2010). Minimizing heat generation in solid-state lasers, IEEE J.
Quantum Electron., 46(7), pp. 1076–1085.
11. Bowman, S. R., O’Connor, S., Biswal, S. (2005). Ytterbium laser with
reduced thermal heating, IEEE J. Quantum Electron., 41(12), pp. 1510–
1517.
12. Bowman, S. R., Jenkins, N. W., O’Connor, S., Feldman, B. J. (2002).
Sensitivity and stability of a radiation balanced laser system, IEEE J.
Quantum Electron., 38(10), pp. 1339–1348.
13. Bowman, S. R., Mungan, C. (2000). New materials for optical cooling,
Appl. Phys. B, 71, pp. 807–811.
14. Bowman, S. R. (1999). Lasers without internal heat generation, IEEE J.
Quantum Electron., 35, pp. 115–122.
15. Bowman, S. R., Quarles, G. J., Lynn, J. G., Feldman, B. J. (1993). A
comparative study of diode pumped two micron laser materials, OSA
Proc. Adv. Solid-State Lasers, 15, pp. 415–418.
16. Bowman, S. R., Feldman, B. J. (1992). Demonstration and analysis of a
holmium quasi-two level laser, Proc. SPIE, 1627, p. 46.
17. Brignon, A., Huignard, J. (2004). Phase Conjugate Optics (John Wiley &
Sons).
18. Budni, P. A., Lemons, M. L., Mosto, J. R., Chicklis, E. P. (2000). High-
power/high-brightness diode-pumped 1.9 μm thulium and resonantly
pumped 2.1 μm holmium lasers, IEEE J. Sel. Top. Quantum Electron.,
6(4), pp. 649–635.
19. Byer, R. L., Basu, S. (1989). Moving slab laser, US Patent 4833682.
20. Chernoch, J. P., Martin, W. S. (1972). Multiple internal reflection face-
pumped laser, US Patent 3633126.
21. Chernoch, J. P., Koenig, H. R. (1969). Disk laser having pumping means
in direct optical combination with the disk end faces, US Patent
3423693.
22. Condon, N. J., Bowman, S. R., O’Connor, S. P., Myers, M. J. (2010). Heat
loads in erbium-doped laser materials, Opt. Mater., 32, pp. 1050–1054.
23. Condon, N. J., Bowman, S. R., O’Connor, S. P., Quimby, R. S., Mungan, C.
5472.
References 185
24. Creeden, D., Johnson, B. R., Setzler, S. D., Chicklis, E. P. (2014). Resonantly
pumped Tm-doped fiber laser with >90% slope efficiency, Opt. Lett.,
39, pp. 470–473.
25. Eggleston, J., Kane, T., Kuhn, K., Unternahrer, J., Byer, R. L. (1984). The
slab geometry laser 1. Theory, IEEE J. Quantum Electron., 20(3), pp.
289–301.
26. Fan, T. Y., Ripin, D. J., Aggarwal, R. L., Ochoa, J. R., Chann, B. Tilleman, M.,
Spitzberg, J. (2007). Cryogenic Yb3+-doped solid-state lasers, IEEE J.
Select. Top. Quantum Electron., 13(3), pp. 448–458.
27. Fan, T. Y. (1992). Optimizing the efficiency and stored energy in quasi-
three-level lasers, IEEE J. Quantum Electron., 28, pp. 2692–2697.
cooling in bulk erbium-doped materials, Phys. Rev. Lett., 97, pp.
0330011–0330014.
29. Fowler, W. B., Dexter, D. L. (1962). Relation between absorption and
emission probabilities in luminescent centers in ionic solids, Phys. Rev.,
128(5), pp. 2154–2165.
30. Friebele, E. J. , Askins, C. G., Peele, J. R., Wright, B. M., Condon, N. J.,
O’Connor, S., Brown, C. G., Bowman, S. R. (2014). Ho-doped fiber
for high energy laser applications, Proc. SPIE, 8961, pp. 8961201–
8961209.
31. Gapontsev, D. (2008). Recent progress on high power fiber lasers, in
11th Annual Directed Energy Symposium, DEPS 1–27.
32. Garbuzova, D., Kudryashov, D. I., Dubinskii, M. (2005). 110 W, 0.9 J
pulsed power from resonantly diode-laser-pumped 1.6-μm Er:YAG
laser, Appl. Phys. Lett., 87, pp. 1211011–1211013.
33. Giesen, A., Speiser, J. (2007). Fifteen years of work on thin-disk lasers:
results and scaling laws, IEEE J. Sel. Top. Quantum Electron., 13(3), p.
598.
34. Giesen, A., Hugel, H., Voss, A., Witting, K., Brauch, U., Opower, H. (1994).
Scalable concept for diode-pumped high-power solid-state lasers,
Appl. Phys. B, 58, pp. 365–372.
35. Guhur, A., Jackson, S. D. (2010). Efficient holmium-doped fluoride
fiber laser emitting 2.1 μm and blue upconversion fluorescence upon
excitation at 2 μm, Opt. Express, 18(19), pp. 20164–20169.
36. Hehlen, M. P., Epstein, R. I., Inoue, H. (2007). Model of laser cooling
in the Yb3+-doped fluorozirconate glass ZBLAN, Phys. Rev. B, 75, pp.
144302–144313.
37. Hemming, A., Simakov, N., Davidson, A., Bennetts, S., Hughes, M.,
Carmody, N., Davies, P., Corena, L., Stepanov, P., Haub, J., Swain, R.,
Carter, A. (2013). Monolithic cladding pumped holmium-doped fibre
laser, IEEE Conference on Lasers and Electro-Optics, CW1M.1.
E. (2000). Observation of anti-Stokes fluorescence cooling in thulium-
39. Injeyan, H., Goodno, R. (2011). High Power Laser Handbook (McGraw-
Hill).
40. Jauregui, C., Eidam, T., Limpert, J., Tunnermann, A. (2011). The
impact of modal interference on the beam quality of high-power fiber
amplifiers, Opt. Express, 19, pp. 3258–3271.
41. Kane, T. J., Eggleston, J. M., Byer, R. L. (1985). The slab geometry laser:
part II; thermal effects in a finite slab, IEEE J. Quantum Electron., QE-
21, pp. 1195–1210.
42. Kim, W., Bowman, S. R., Baker, C., Florea, C., Villalobos, G., Shaw, B.,
Sadowski, B., Hunt, M., Aggarwal, I., Sanghera, J. (2014). Holmium
doped laser materials for eye-safe solid state laser application, Proc.
SPIE, 90814, pp. 9081051–9081056.
43. Klopp, P., Petrov, V., Griebner, U. (2003). Potassium ytterbium tungstate
provides the smallest laser quantum defect, Jpn. J. Appl. Phys., 42, pp.
246–248.
44. Lacovara, P., Wang, C. A., Choi, H. K., Aggarwal, R. L., Fan, T. Y. (1991).
Room-temperature InGaAs diode-pumped Yb-YAG laser, OSA Proc. Adv.
Solid-State Lasers, 10, pp. 283–285.
45. Larat, C., Schwarz, M., Lallier, E., Durand, E. (2014). 120 mJ Q-switched
Er:YAG laser at 1645 nm, Opt. Express, 22(5), pp. 4861–4866.
46. Li, C., Liu, Q., Gong, M., Chen, G., Yan, P. (2004). Modeling of radiation-
balanced continuous-wave laser oscillators, J. Opt. Soc. Am. B, 21(3),
pp. 539–542.
47. Loiko, P., Maria Serres, J., Mateos, X., Yumashev, K., Kuleshov, N.,
Petrov, V., Griebner, U., Aguiló, M., Díaz, F. (2015). In-band-pumped
Ho:KLu(WO4)2 microchip laser with 84% slope efficiency, Opt. Lett.,
40(3), pp. 344–347.
48. Mandal, A., Klimek, D. E. (2010). Textron’s J-HPSSL 100 kW ThinZag,
IEEE Conference on Lasers and Electro-Optics, JThH2.
49. Marmo, J., Injeyan, H., Komine, H., McNaught, S., Machan, J., Sollee, J.
(2009). Joint high power solid state laser program advancements at
Northrop Grumman, Proc. SPIE, 7195, pp. 7195071–7195076.
References 187
50. McCumber, D. E. (1964). Einstein Relations connecting broadband

emission and absorption spectra, Phys. Rev., 136(4A), pp. 954–957.
51. Nemova G., Kashyap, R. (2009). Athermal continuous-wave fiber
amplifier, Opt. Commun., 282, pp. 2571–2575.
52. Newell, T. C., Peterson, P., Gavrielides, A., Sharma, M. P. (2007).
Temperature effects on the emission properties of Yb-doped optical
fibers, Optics Commun., 273, pp. 256–259.
53. O’Connor, S. P., Biswal, S., Condon, N., Bowman, S. R. (2011).
Demonstration and analysis of a high power radiation balanced laser,
IEEE Conference on Lasers and Electro-Optics, CMH4.
54. Oulundsen, G., Carter, A., Samson, B., Bowman, S. R. (2012). Holmium-
doped silica fiber designs extend fiber lasers beyond 2 µm, Laser Focus
World, 48(8), pp. 47–50.
55. Petermann, K., Huber, G., Fornasiero, L., Kuch, S., Mix, E., Peters, V.,
Basun, S. A. (2000). Rare earth–doped sesquioxides, J. Lumin., 87, pp.
973–975.
56. Petit, J., Viana, B., Goldner, P., Vivien, D., Louiseau P., Ferrand, B. (2004).
Laser oscillation with low quantum defect in Yb:GdVO4, a crystal with
high thermal conductivity, Opt. Lett., 29(8), pp. 833–835.
57. Pollnau, M., Hardman, P. J., Kern, M. A., Clarkson, W. A., Hanna, D. C.
(1998). Upconversion-induced heat generation and thermal lensing in
Nd:YLF and Nd:YAG, Phys. Rev. B, 58(24), pp. 16076–16091.
58. Quimby, R. S., Condon, N. J., O’Connor, S. P., Biswal, S., Bowman, S. R.
(2008). Upconversion and excited-state absorption in the lower levels
of Er:KPb2Cl5, Opt. Mater., 30(6), pp. 827–834.
59. Richardson, D. J., Nilsson, J., Clarkson, W. A. (2010). High power fiber
lasers: current status and future perspectives, J. Opt. Soc. Am. B, 27(11),
pp. 63–92.
60. Sanghera, J., Kim, W., Villalobos, G., Shaw, B., Baker, C., Frantz, J.,
Sadowski, B., Aggarwal, I. (2012). Ceramic laser materials, Materials,
5, pp. 258–277.
61. Shen, D. Y., Sahu, J. K., Clarkson, W. A. (2006). Highly efficient in-band
pumped Er:YAG laser with 60 W of output at 1645 nm, Opt. Lett., 31,
pp. 754–756.
62. Smith, A. V., Smith, J. J. (2011). Mode instability in high power fiber
amplifiers, Opt. Express., 19, pp. 10180–10192.
63. Stoneman, R. C., Esterowitz, L. (1992). Intracavity-pumped 2.09-,um
Ho:YAG laser, Opt. Lett., 17(10), pp. 736–738.
64. Takaichi, K., Yagi, H., Shirakawa, A., Ueda, K., Hosokawa, S., Yanagitani,
T., Kaminskii, A. A. (2004). Lu2O3:Yb3+ceramics a novel gain material
for high-power solid-state lasers, Phys. Status Solidi A, 202, pp. R1–R3.
65. Ward, B., Robin, C., Dajani, I. (2012). Origin of thermal modal
instabilities in large mode area fiber amplifiers, Opt. Express, 20, pp.
11407–11422.
66. White, J. O., Dubinskii, M., Merkle, L. D., Kudryashov, I., Garbuzov, D.
(2007). Resonant pumping and upconversion in 1.6 μm Er3+ lasers, J.
Opt. Soc. Am. B, 24(9), pp. 2454–2460.
67. Wright, J. (1976). Up-conversion and excited state energy transfer in
rare-earth doped materials, Radiationless Processes in Molecules and
Condensed Phases, Vol. 15 of Topics in Applied Physics, Chapter 4, pp.
239–295.
68. Yao, T., Ji, J., Nilsson, J. (2014). Ultra-low quantum-defect heating in
ytterbium-doped aluminosilicate fibers, J. Lightwave Tech., 32(3), pp.
429–434.
Chapter 6
Methods for Laser Cooling of Solids
Stephen C. Rand
MURI Center for Dynamic Magneto-Optics, Appl. Physics & EECS Optics & Photonics
Laboratory, 3102 ERB1, 2200 Bonisteel Blvd., University of Michigan, MI, USA
scr@umich.edu
6.1 Introduction
The topic of laser cooling began with the articles of Pringsheim
[1] and Hansch [2] long before any laboratory experiments were
undertaken. These early papers proposed the use of optical radiation
to remove energy from matter for the purpose of refrigeration, but
the mere idea that light could transport thermal energy was widely
doubted until Landau’s thermodynamic analysis showed that
radiation fields had entropy that depended on the statistics and
mode structure of the field [3]. The influence of various states of the
field on solid state cooling was examined in detail later [4], but once
it was realized that entropy of fields could be exchanged for entropy
of matter, the stage was already set for a concerted worldwide effort
to alter the temperature of collections of atoms with light.
Experiments to observe laser cooling of atomic motion in
gases originated from investigations of the mechanical effects of

www.panstanford.com
190 Methods for Laser Cooling of Solids
radiation pressure on small dielectric particles [5]. Similarly, optical

refrigeration of solids was preceded by demonstrations that the
amplitude of axial vibrations of mirrors could be reduced in Fabry–
Perot resonators [6]. Reductions of this kind were accomplished
by monitoring excursions of the cavity from resonance and using
the error signal to control mirror position with variations of light
pressure applied to the back side of one of the cavity mirrors.
However, this type of optomechanical feedback does not lead to
overall refrigeration, because it does not reduce the amplitude of
internal modes of vibration irreversibly. Some modes are heated,
while others are cooled. To refrigerate a solid, light must do more
than redistribute vibrational energy among the available modes.
Ideally a low-entropy (single-mode) input field couples to one or
more vibrational or translational modes, undergoes an increase in
entropy during its interaction with the sample, and then carries that
entropy away via radiative transport into space.
Laser cooling techniques for solids can build on past experience
in atomic vapors. In gases the motion of atoms along a chosen
direction can be exploited to shift “hot” atoms preferentially into
resonance with light propagating opposite to their motion via the
Doppler effect [7]. In this way the momentum of the light wave is
transferred to fast-moving atoms selectively, gradually slowing them
down through the accumulated effect of many opposing photon
interactions. In condensed matter, however, free translational motion
is absent. As a consequence, one might think that laser cooling in
glasses or crystals has to exploit completely different principles.
Yet there are analogies to the Doppler effect in solids and ways to
address the energy and momentum of collective excitations of the
medium that are reminiscent of interactions with free atoms. Hence
this chapter reviews several methods with this perspective in mind,
attempting to profit from the earlier history of laser cooling in gases
and to provide points of comparison between different techniques.
Researchers face the challenge of developing the most effective
means of controlling the temperature of macroscopic objects
with light. Success on this front will make possible the thermal
management of hot spots in micromechanical or integrated
electronic devices, improved operation of high-power lasers through
radiation balancing, and low-temperature operation of active devices
Cooling with Anti-Stokes Fluorescence 191
in vacuum where cooling improves sensitivity but is only possible

through radiative transport.
6.2 Cooling with Anti-Stokes Fluorescence

The earliest demonstration of optical refrigeration of a solid relied on
inducing anti-Stokes fluorescence by simply tuning the wavelength
of incident light to the “red” or low-energy side of an impurity
resonance [8]. When vibronic coupling of the medium is weak, the
detuning is into the spectral wing of an essentially pure electronic
transition. When the vibronic coupling is strong, a condition
characterized by a large Huang–Rhys factor [9], the light is detuned
into the phonon sideband associated with the electronic transition.
Then, consistent with quantum theory [10], the absorption of
photons on the low-energy side of resonance is followed on average
by photon emission at the center frequency of the transition. In
this circumstance an overall loss of energy is imposed even on
stationary atoms (Fig. 6.1). Hence this general approach has met
with considerable success in demonstrations of cooling in rare earth
solids [1], liquids [12], semiconductors [13], and internal acoustic
modes of optomechanical resonators [14].
The efficiency of cooling that relies on inducing anti-Stokes
fluorescence can easily be estimated. In the two-level system of Fig.
6.1, an amount of energy given by hv is added to the system for each
absorbed photon. Energy hvfi is removed by each fluorescent photon.
Hence if these two steps account for all system dynamics, then an
input power of Pin produces a cooling power per absorbed photon of
Ê n - nL ˆ
Pc = Pin Á fl (6.1)
Ë n L ˜¯
and the cooling efficiency would be η = Pc/Pin = (vfi – vL)vL. However,
in the presence of nonradiative relaxation, which is commonplace
in rare earth solids, the radiative quantum efficiency is lowered.
Radiative and nonradiative decay at rates Wr and Wnr then contribute
an external efficiency factor of the form ηext = Wr/(Wr + Wnr). When
corrected for fluorescence escape efficiency ηe, which is the fraction
of emitted photons that escape the sample, this factor becomes
ηext = ηrWr/(ηeWr + Wnr) (6.2)
ν νfl
1
Figure 6.1 Anti-Stokes fluorescence emission in a two-level system.
Additionally, only a fraction of the absorbed photons result in

excitation of coolant atoms. The rest are absorbed by background
impurities. With the absorption coefficient of the coolant atoms
designated by αc and that of the background by αb, the proportion of
absorbed photons that contribute to cooling is
ac
habs = (6.3)
(a c + a b )
If the assumption is made that phonon-mediated absorp-
tion is proportional to the occupation probability of the phonon
mode, namely n = (exp[W/kBT ]-1)-1, the efficiency becomes
habs = a c n /(a c n + a b ). Then at high temperatures (kBT >> ħΩ)),
ηabs varies little with temperature when the background absorp-
tion is low (αb/αc << 1). On the other hand at low temperatures, as
n Æ 0, ηabs decreases in proportion to n . Hence the absorption
efficiency drops dramatically through the important cryogenic re-
gime. The overall cooling is determined by the product of the various
efficiency factors.
Ê n -n ˆ
hc = hext habs Á fl (6.4)
Ë n ˜¯
From Eqs. 6.1 and 6.4 it is easy to show that the cooling power
density per unit frequency is Pc/V = ηabsηextαtotIv(vfi – v)/v. Here Iv
= I0g(v) is the intensity per unit frequency and g(v) is a normalized
spectral line shape function. V is the interaction volume. If the
bandwidth of the light source equals the full linewidth 2Γ of the
optical transition, and the density of atoms in the interaction volume
is N/V, the energy loss rate per atom is
Cooling with Anti-Stokes Fluorescence 193
R = 2ΓηcαtotIv(V/N) = ηcαtotI0(V/N) (6.5)

The condition for net cooling can be expressed in a simple way
based on Eq. 6.4, since ηc must exceed zero. By assuming that the
laser detuning is optimized at the phonon frequency Ω, we can
rewrite the numerator using hvft – hv = ħΩ. Furthermore the average
phonon energy is ħΩ ≈ kBT near the boundary of the classical regime
[15]. Then the condition for laser cooling (ηc > 0) becomes
kBT
hext habs > 1 - (6.6)
hn
An important limitation of refrigeration based on anti-Stokes
fluorescence emerges from this condition by determining the
temperature at which the cooling efficiency drops to zero (ηc = 0).
By solving Eq. 6.6 as an equality, it is found that a lower bound exists
for the attainable temperature. The minimum temperature Tmin that
can be reached with anti-Stokes cooling found from Eq. 6.6 is
hn Ê (a b / a c ) + (1 - hext ) ˆ
Tmin = ˜¯ (6.7)
kB ÁË 1 + (a b / a c )
According to Eq. 6.7, if the quantum efficiency is high (next ≈ 1),
the temperature limit for laser cooling by anti-Stokes emission is
primarily determined by the frequency of light and the background-
to-resonant-absorption ratio. The dependence on these two factors
arises because heat input to the system is proportional to incident
photon energy and absorptive heating efficiency. Only when the
background absorption coefficient vanishes could arbitrarily low
temperatures be reached. For the moderate levels of background
absorption encountered in practice in solids, this places a severe
constraint on temperatures that can be reached by laser cooling.
As a simple example of this, the anti-Stokes cooling limit predicted
by Eq. 6.7 at a wavelength of 1 micron is Tmin = 145.3 K when the
external quantum efficiency is ηext = 0.90 and background absorption
coefficient is αb = αc/1000.
The lower bound on attainable temperatures can be interpreted
in another way. It arises because the cooling rate depends on phonon
occupation, a factor that diminishes as temperature decreases.
Eventually the rate of anti-Stokes cooling drops below the rate of
heating from background impurity absorption and cooling is no
longer possible [16]. For 3D refrigeration of solids to be as effective as
laser cooling of gases, techniques capable of maintaining high rates

of cooling through the cryogenic temperature range are needed. This
is discussed further in what follows.
In the remainder of this chapter momentum is incorporated
into the discussion to investigate whether the fundamental
temperature limitation of anti-Stokes fluorescence cooling can be
evaded when the dynamics are viewed comprehensively. Judging
from the successes of laser cooling of gases, cooling schemes that
conserve momentum and energy simultaneously during optical
interactions are the most efficient. Yet momentum is disregarded
in the anti-Stokes fluorescence method. Other techniques may offer
the capability of balancing the momentum and energy requirements
more consistently to accommodate phonon dispersion (Fig. 6.2).
q→
+ _
_
+ + +
_ _ T
+
Ω q→
_ +
_
+ T+ +
_ _
+ q
π /a
Figure 6.2 (Left) Dispersion curves illustrating the variation of phonon
frequency W with wavevector q for transverse acoustic (TA)
and transverse optic (TO) modes. The lattice constant is a.
(Right) Particle displacements in a diatomic linear lattice for
TO (upper) and TA (lower) modes of the same wavelength.
6.3 Brillouin Cooling

In gases the link between linear momentum and the absorption of a
photon is simple. For each photon absorbed, the linear momentum
of an approaching atom decreases by an average of ħk. Translational
momentum is thereby reduced, provided fluorescence randomizes
the direction of reemitted photons. In solids things are not so
Brillouin Cooling 195
simple because free translation of atoms cannot take place. Atomic

displacements are oscillatory. Motion is conveyed in collective waves
and momentum is distributed, as illustrated in Fig. 6.2. Nevertheless,
charge motions associated with departures of atoms from their
equilibrium positions have Fourier components with a well-defined
amplitude, momentum, and energy [11]. Moreover, Doppler-shifted
interactions with these phonon modes are well-known to take
place in acousto-optic (Brillouin) scattering [15]. So analogies
to Doppler cooling methods in gases are useful. At this point we
therefore formulate the problem of light interacting with acoustic
and optical modes of vibration in terms of interaction Hamiltonians
and equations of motion that display momentum explicitly. Two
recent developments are reviewed. The most detailed consideration
is given to laser cooling by stimulated Raman scattering because in
theory it furnishes a cooling rate that is independent of temperature.
The energy of interaction between light and a solid derives
from the electron–phonon interaction in which the force of light
acts to displace ions from equilibrium by an amount δR*(t). This
displacement can be written as a Fourier expansion in terms
of normal modes of polarization *εkλ (λ = 1,2,3), amplitude Qqλ,
wavevector q, and frequency Ωqλ. For an ion situated at equilibrium
position R0 in a solid, undergoing motion described by a single-
phonon mode, the displacement consists of only one Fourier term.
d R(t ) = ( NM )-1/2 Q(W , t )e exp[iq ◊ R0 ] (6.8)
e is the polarization vector, N is the number of particles, and M is
the particle mass. The form of the interaction Hamiltonian is given
by force × distance.
HI = -—V ◊d R (6.9)
The quantized form of HI depends on the coupling mechanism
between light and the vibrational mode of interest, so the cases of
acoustic and optical vibrational modes are considered separately
below.
For transverse acoustic modes, which are propagating density
fluctuations, the light–matter interaction is determined by
electrostrictive deformations of the medium because transverse
optical fields cannot couple directly to longitudinal waves. The
energy dependence of acoustic and optical modes on the wavevector
is shown in Fig. 6.2. To generate pressure waves, a deformation

potential [17] is needed of the form
—V = -g—(E ◊ E * ) = -g qE E * e (6.10)
1 2 1 2
Here
- +
E1 ( z , t ) = xk (a 1 e - i(w1t - kz ) + a 1 ei(w1t - kz ) ) (6.11)
and
+ -
E2* ( z , t ) = xk + q (a 2 ei(w2t -[ k + q]z ) + a 2 e - i(w2t -[ k +q ]z ) ) (6.12)
are quantized pump and anti-Stokes fields, xk = w k / 2e0V is the

field per photon, and γ = [ρ(∂ε /∂ρ]ρ0 is the electrostrictive constant
relating changes of permittivity to changes of density [18]. The
operator for normalized mode amplitude is related to the phonon
+ -
creation ( b ) and annihilation ( b ) operators by
1/ 2
(W , t ) = Ê ˆ
Q
- +
(b q exp[ - iWt ] + b q exp[iWt ]) (6.13)
ÁË 2W ˜¯
When Eqs. 6.8 and 6.10–6.13 are combined with Eq. 6.9 the
interaction Hamiltonian becomes
H I = -( NM )-1/2 (e ◊ —V )Q(W , t )exp[iq ◊ R0 ]
- -
= f (b q ak++ q a k + h.c .) (6.14)
where the interaction strength is defined as f *₌ (2ħNMΩ)–1/2Vγqξkξk+q.
V is the volume of interaction. The conservation of momentum in the
interaction is illustrated in Fig. 6.3.
Ω, q
ω2 , k2
θ
ω1 , k1
Figure 6.3 Wavevector diagram for phonon scattering through angle q.

Phonon frequency W and wavevector q = k2 - k1 conserve
energy and linear momentum in the wave interaction involving
optical fields E1 (w1 ) and E2 (w 2 ).
Brillouin Cooling 197
The Hamiltonian can be used to calculate temporal changes in the

vibrational energy (or temperature) of the medium. To appreciate
the connection between HI and vibrational energy or temperature,
note that the conjugate momentum of the vibrational amplitude
(W , t ) is
Q
1/ 2
Ê W ˆ
P (W , t ) = -i Á (bexp[ - iWt ] - b exp[iWt ]) (6.15)
+

Ë 2 ˜¯
When Q and P are defined as above, the single-mode
Hamiltonian specifying vibrational energy reduces to a familiar
form.
+
+ Q
Hql = 12 ( P ql P ql + W2ql Q + -
ql ql ) = 2 Wql ( bql bql + h.c .) (6.16)
1
Taking all possible modes into account, the total energy is

therefore found to be
H= ÂH
q ,l
ql = Â W
q ,l
+ -
ql ( bql bql + 12 ) (6.17)
+ -
Since n ql = b ql b ql is the number of phonons in mode q, the
summation in Eq. 6.17 describes the total vibrational energy of the
medium as a weighted sum of the individual mode energies which in
turn determines the sample temperature.
With the interaction Hamiltonian of Eq. 6.14 in hand,
optically driven dynamics of two optical waves interacting
with an acoustic mode can be analyzed. The static Hamiltonian
+ - + - + -
H = w a 1 a 1 + w a 2 a 2 + Wb b plays no role in steady-state
0 1 2
dynamics. For example, the change of phonon occupation with time
can be calculated using the Heisenberg equation of motion
.
n b = (i / )[H0 + Hint , n ]
- + -
{ + -
= (i / ) f (b a 2 a 1 + h.c .), b b
}

- + - + -
{ + - + - +
= if b a 2 a 1 (b b ) - b (b b )a 2 a 1 (6.18) }
The first term on the right determines the rate of annihilation
of phonons, while the second term is the rate of creation. We can
+
show this explicitly by solving for the temporal behavior of b and
-
b , again using the Heisenberg picture.
+ + - - + + + + - - +
(db /dt ) = iWb + if b a 1 a 2 b - if b a 1 a 2 b - G b b (6.19)
- - + + - - - - + + -
(db / dt ) = -iWb - if b a 1 a 2 b + if b a 1 a 2 b - G b b (6.20)
Making use of the slowly varying envelope approximation (SVEA),
steady-state solutions for the operator amplitudes are simply

+
{ - - + + + + - -
}
b = if b a 1 a 2 b - if b a 1 a 2 b / G b

-
{ + + - - - - + +
}
b = - if b a 1 a 2 b + if b a 1 a 2 b / G b (6.21)
With these substitutions one finds that the net rate of change
of phonon occupation in the state n1 n2 nb is determined by the
balance of annihilation versus creation.
{ }
.
+ + - - - - + + + - + -
n b = if nb n1 n2 - if b a 1 a 2 b + if b a 1 b 2 b a 2 b 1 (b b ) n2 n1 nb / G b
{
- - + + + + - -
}
- + - +
- if nb n1 n2 if b a 1 a 2 b - if b a 1 a 2 b a 2 a 1 (b b ) n2 n1 nb / G b
(6.22)
Upon evaluation of the matrix elements we obtain the rate
equation for phonon occupation.
nb = -( f 2/G b )n1 (n2 + 1)nb + ( f 2/G b )(n1 + 1)n2 (nb + 1) (6.23)
The first term on the right side of Eq. 6.23 describes cooling,
while the second describes heating. Possible detuning of the modes
has been ignored in this analysis. The phonon annihilation rate for
resonant Brillouin scattering in the interaction volume is therefore
the coefficient of the first term in Eq. 6.23.
γB = (f2/Γb)n1(n2 + 1)nb (6.24)
The energy loss rate takes into account the phonon frequency.
R = γBħΩ = (f2/Гb)n1(n2 + 1)nbħΩ (6.25)
Laser cooling is realized if conversion of an optical pump wave
to an anti-Stokes wave prevails over its conversion to a Stokes-
shifted wave [19]. Brillouin cooling of a mechanical mode was first
demonstrated in a microresonator by simultaneously matching
Raman Cooling 199
its momentum and energy to the difference of wavevectors and

frequencies of the pump and anti-Stokes waves [14]. To do this the
pump and anti-Stokes waves were simultaneously enhanced in the
resonator, while the Stokes component of Brillouin scattering was
suppressed by wavevector mismatch.
6.4 Raman Cooling

In this section stimulated scattering is considered for the cooling
of solids. The general theory of stimulated Brillouin and Raman
scattering [20, 21] was developed in the 1960s, and despite the
complexity of equations coupling the various mode amplitudes,
Louisell [22] pointed out that the interaction energy is just the added
electromagnetic energy resulting from a change in the permittivity
Ú
ε, or HI = 12 deE 2. This perspective facilitates the analysis of both
Brillouin and Raman scattering since it is straightforward to write δε
to reflect one interaction or the other. In the last section the dielectric
fluctuation in Brillouin scattering was written δε = –l(∂Q/∂x)to
describe electrostriction by the incident light. In this section δε
arises from dielectric fluctuations due to transverse charge motion
caused directly by the light field itself. We shall proceed as before
nevertheless, writing down the interaction Hamiltonian using
Eq. 6.9, since the deformation potential approach is equivalent to
that of Louisell.
For transverse optical modes, the net optical electric field can
act directly on charged ions at the difference frequency to create
or destroy optical phonons. In this case the potential is just the
optical energy density at the mode frequency. Hence the gradient of
the potential in Eq. 6.9 is —V = e —(E ◊ E * )e , and the single-mode
0 1 2
interaction Hamiltonian is
1/ 2
Ê e02 ˆ - + -
HI = Á ˜ xk xk + q (a 1 a 2 b + h.c .)
Ë 2WNM ¯
- + -
= f '(a 1 a 2 b + h.c .) (6.26)

where f ´ = –(2ħNMΩ)–1/2ε0Vq ξk ξk+q is the coupling strength in the
interaction volume V. By careful selection of the frequencies and
wavevectors of two optical fields, the energy and momentum can be

conserved for optical modes just as for acoustic modes (Fig. 6.3).
Two-photon Raman transitions (Fig. 6.4) can couple to
optical modes of vibration purely through anti-Stokes scattering.
Consequently they can be utilized for efficient cooling. Early analysis
and experiments on Raman cooling may be found in Refs. [23–25].
Here we follow Ref. [26] to analyze how the Raman technique can
be adapted to cooling solids and to estimate the cooling rate. The
potential advantages of applying adiabatic rapid passage to this
technique are also considered. The fact that the cooling rate is
temperature independent is emphasized.
With pulsed excitation, the two fields at ω1 and ω2 in Fig. 6.4
cause the system population to evolve in the direction indicated by
wiggly arrows. The light drives population from state 1 to state 3 at
a rate determined by the one- and two-photon detunings ∆ and δ.
The 1 Æ 2 and 2 Æ 3 transitions are assumed to be electric-dipole-
allowed. The selection rules therefore dictate that state 3 is a long-
lived storage state. For prolonged excitation, population builds up in
level 3 and the reverse process begins to take place at a comparable
rate. For the two-photon detuning shown in the figure, a small
energy deficit (ħδ) and momentum reduction (2ħk) is incurred for
each Raman cycle, provided the pulses are short enough to avoid
substantial buildup of population [27] in state 3. Atoms with a
velocity centered at v = 2ħk/m lose kinetic energy. By emptying state
3 periodically with a fast, random process like fluorescence, changes
in the energy and momentum distributions become irreversible
through an increase in the entropy [3, 4] of the outgoing radiation
field [28]. Hence the ensemble of atoms can be progressively cooled.
To estimate the stimulated Raman transition rate for the 1 Æ 3
transition it is instructive to develop a sense of the importance of
forward and backward pumping processes between states 1 and 3
(Figs. 6.4 and 6.5, respectively). It is easily shown that the net rate
of change of population is zero in the steady state because of the
balance of these two processes, which provide cooling and heating,
respectively. One can also show that the transition rate for a single
process is given by the square of an off-diagonal (coherence) element
of the density matrix that may be evaluated straightforwardly to
allow estimation of cooling rates.
Raman Cooling 201
2
∆
ω1 ω2
δ
3
1
Figure 6.4 A three-level system undergoing a two-photon Raman process.
∆ =ω12 – ω is the one-photon detuning and δ =ω13 – (ω1 – ω2)
is the two-photon detuning.
2
∆
ω1 ω2
δ
3
1
Figure 6.5 Raman interaction of fields E1(ω1) and E2(ω2) in a time-
reversed (backward) sequence.
The net rate of population change in excited state 3 is just

r33 ∫ d r33 /dt . In the weak field limit the wave function is dominated
by its ground-state component, which varies little with time. Hence
we may assume r11 = c1c1* @ 1 in order to calculate the transition rate
d d
È 3 y y 3 ˘˚ @ È 3 c3 3 1 c1*c1 1 3 c3* 3 ˘ = r31 r13 + r31 r13
r33 =
dt Î dt Î ˚
(6.27)
Since r31 = r13 *

and |ρ31| = |ρ13|, one finds r31 = r13
*
= -iw13 r31
and r13 = iw13 r13. The result is that the two contributions to ρ33 given
in Eq. 6.27 are equal and opposite. Under steady-state conditions the
occupation of state 3 therefore does not change.
2 2
r33 = -iw13 r13 + iw13 r13 = 0 (6.28)
The rate at which the level 3 population increases is determined

by the positive term in Eq. 6.28, assuming one vibrational quantum
occupies the mode. The transition rate per atom per unit (angular
frequency) excitation bandwidth that leads to cooling is therefore
2
r33 / w13 = r13 (6.29)
To determine the rate itself from the field-driven coherence in
Eq. 6.29 the source bandwidth must be specified. Taking this to
equal the full transition linewidth 2 Г13 in the discussion below, only
the calculation of the matrix element ρ13 explicitly remains.
For quasi-, steady-state fields the interaction Hamiltonian can be
written as
V = ( - 12 W12exp(iw1t ) + c .c .) + ( - 12 W23exp(iw 2t ) + c .c .) (6.30)
Contributions to state 3 dynamics other than the one depicted
in Fig. 6.4 will be ignored, owing to their large detunings from
resonance. The first-order coherence is therefore given by the
standard first-order perturbation expression [29]:
(1) Ê W12 / 2 ˆ
r12 =Á ( r11
(0)
- r22
(0)
) (6.31)
Ë D1 + iG12 ˜¯
The second-order coherence established by the forward Raman
process is obtained by solving the second-order transport equation
i r 13
( 2)
= (w1 - w3 )r13
( 2)
+ [V (1) , r(1) ]13 - i G 13 r13
(2)
(6.32)
The two-photon coherence oscillates at the frequency of
the (combined) driving fields. So if the slowly varying envelope
approximation is adopted, the form of ρ13 is
r13
( 2)
(t ) = r13
( 2)
exp[i(w1 - w 2 )t ] (6.33)
By substituting Eqs. 6.30, 6.31, and 6.33 into Eq. 6.32 and
grouping terms with the same time dependence, one finds the
solution for the coherence to be
Raman Cooling 203
ÏÔÊ W12W23 * ˆ (0) ¸

/4 (0) Ô
r13
( 2)
= ÌÁ ˜ ( r11 - r22 )˝ (6.34)
[ D + iG 12 ][ D 2 + iG 13 ]¯
ÓÔË 1 ˛Ô
The second denominator incorporates the two-photon detuning
∆2 = ω31 – (ω1 – ω2). Using this result in Eq. 6.29, the Raman transition
rate per atom is obtained with an appropriate assumption regarding
excitation bandwidth. If the excitation matches the bandwidth of the
Raman transition (2Г13), the number of phonon annihilations per
impurity atom per second is
2 2
g R = 2G 13 < n > r 13
( 2)
= 2G 13 < n > f M i (6.35)
where adjustment has been made for the number of phonons occupy-
ing the mode using the Planck distribution n = (exp[W / kBT ] - 1)-1 .
In this model the implicit summation over intermediate states in the
matrix element f M i reduces to a single term:
È 3 m (e ) ◊ e 2 2 2 m (e ) ◊ e 1 1 ˘ E(w )E *( -w )
f M i =Í ˙ 1 2
(6.36)
ÍÎ [ D1 + iG 12 ][ D 2 + iG 13 ] ˙˚ 4 2
The rate of energy loss is therefore
2
R = g R W = 2G 13 n f Mi W (6.37)
This rate applies to each atom in the interaction volume because
it was calculated using the density matrix that is normalized to the
number of coolant ions. To realize this cooling rate in practice, the
optical interaction must be made irreversible. To this point, the
dynamics we have described are coherent and proceed as readily in
the forward direction as in the backward direction. We must ensure
that the second term in Eq. 6.27 describing reverse transitions is
negligible. Also the overall efficiency may benefit from interaction
with more than one mode at a time. So we briefly consider these
aspects of Raman cooling next with the aid of Fig. 6.6.
Refrigeration requires irreversibility. The coherent (reversible)
Raman process outlined above must therefore be combined with
a random process that increases the entropy of outgoing radiation
[4] without decreasing efficiency. For this purpose fluorescence on
an allowed transition at frequency ω4 can be induced by a strong
repump beam (at frequency ω3), as illustrated schematically in
Fig. 6.6. To avoid lowering the rate of cooling set by anti-Stokes
Raman transitions, the return of the excited-state population to the
ground state should be fast. Provided state 3 is emptied at a rate

faster than the Raman excitation process that fills it, and no extra
heat is generated in the process, efficient cooling will be maintained.
ω3 ω4
Figure 6.6 The coherent Raman process (dashed arrows) must be

followed by random emission for there to be net cooling. Here,
spontaneous emission is induced at ω4 with a pump source at
frequency ω3 that returns the population from state 3 to state
1.
Improvement in the Raman cooling rate can be anticipated by

interacting with a range of vibrational modes during each cooling
cycle. More than one mode could be addressed, for example, by
rapid scanning the two-photon detuning δ in discrete steps [27] or
by using some form of adiabatic passage. Reversal of the temporal
order of two fixed-frequency pulses can transfer the population
between two discrete states with nearly unit efficiency at a fixed
detuning. However, for rapid cooling over a broad range of phonon
frequencies, rapid adiabatic passage is a superior method in which
the excitation detuning is swept from one side of Raman resonance
to the other adiabatically. This approach can be implemented very
simply by imposing a “chirp” on δ(t). By sweeping the frequency
rapidly from one side of resonance to the other, all phonons with
frequencies between the endpoints ±δmax of the chirp are cooled at
once [30]. Adiabatic passage is discussed further in Chapter 7.
Raman Cooling 205
The need to avoid heat generation during the repump cycle

affects how closely the repump beam may be tuned to resonance.
The criterion for optimizing one-photon detuning in Raman cooling
of solids is different from that for gases because the configuration of
neighboring atoms surrounding a coolant atom changes as the dipole
forms during transition of a coolant atom from a low- to a high-
energy state. That is, the equilibrium positions of neighbors change
in response to the absorption process. Phonons are unavoidably
generated by relaxation of this nonequilibrium configuration unless
the light is detuned on the red side of resonance by more than the
phonon energy.
In the illustration of Fig. 6.7, detuning requirements are
considered in the context of energy levels of Ce3+ ions. Given the
electronic structure of Ce3+, the laser must be detuned by an amount
equal to or greater than the Stokes shift of the longest-wavelength
4fn – 1 5d–4fn transition. For Ce3+ this means
∆Laser ≥ ∆ Stokes + ∆F (6.38)
where ∆F is the energy separating the 2F and 2F energy levels.
5/2 7/2
This choice ensures that absorbed photons have less than the energy
required to generate either a vibrationally excited upper state or a
vibrationally excited ground state. In the Ce3+ system, cooling ought
to maximize when both the Raman and repump beams are detuned
by ∆Laser = ∆Stokes + ∆F. In practice, smaller detunings might improve
the cooling if the transition rate from the 5d state to 2F7/2 is less than
that to 2F5/2, as it is in Ce:YAG [31]. Despite the reduction in cooling
rate that large detunings incur in general, high overall cooling rates
can be achieved when the excited state is exceptionally broad, as
predicted [26] for the 5d state of Ce3+.
To conclude this chapter we emphasize two key advantages of
Raman cooling of solids, namely that the cooling rate is temperature
independent and that the cooling process is divided into two
main steps involving different levels, thereby permitting separate
optimization of absorption and fluorescence. At low temperatures,
the phonon occupation factor in Eq. 6.27 drops below 1 and varies
as <n> = exp(–ħΩ/kBT). In this nonclassical regime, any cooling rate
that is solely proportional to <n> falls exponentially as temperature
decreases. Indeed γR is proportional to <n> according to Eq. 6.35,
just like the rate for anti-Stokes fluorescence cooling. However, the
phonon-broadened linewidth Г13 of the Raman transition is strongly

temperature dependent and offsets the decline with a matching
proportionality to <n> of its own. In the model below there is
complete cancellation of factors of <n> in Eq. 6.35 at two-photon
resonance. Since the transition to state 3 is detuned by the phonon
frequency below the terminal electronic state (see Fig. 6.4), the
relevant resonant condition is ∆2 = 0.
Conducon
4fn-5d-4fn-1 5d-4fn-1 -4fn

absorpon emission
∆Stokes
∆Laser
1 2 3
2
F7/2
∆F
2
F5/2
Figure 6.7 Energy levels and detunings involved in Raman laser cooling of
a three-level coolant atom in a solid. The example is based on
the electronic structure of Ce3+.
While other mechanisms may contribute, we assume here that

the Raman linewidth is governed primarily by a two-phonon Orbach
(
process [32]. Thus it can be modeled as G 13 (T ) @ G 0 1 + n [G '0 / G 0 ] . )
In this expression Г0 is the limiting inhomogeneous linewidth at zero
Raman Cooling 207
temperature. When Г’0 > Г0 the linewidth reduces to G 13 (T ) @< n > G '0.
Upon substitution into Eq. 6.35 with zero two-photon detuning (∆2
= 0), the rate of phonon annihilation per atom becomes constant.
2 2
gR = f M' i (6.39)
G '0
In this expression the modified matrix element M’ appears
because of the cancellation between <n> and two-photon linewidth
factors in the Orbach model outlined above. Its definition is
È 3 m (e ) ◊ e 2 2 2 m (e ) ◊ e 1 1 ˘ E(w )E *( -w )
f M' i = Í ˙ 1 2
(6.40)
ÍÎ [ D1 + iG 12 ] ˙˚ 4 2
The energy loss rate per atom
2 2
R = g R W = f M' i W (6.41)
G '0
is similarly expected to be temperature independent over the entire
range kBT > ħГ0. Unlike acoustic modes, optical phonons have
very little dispersion (Fig. 6.2). Consequently the frequency of the
dominant optical phonon does not vary significantly and R remains
constant over a wide range of temperature.
The 1D Raman interaction described above involves a phonon
near the center of the Brillouin zone propagating along a single optical
axis. However, zone center phonons cannot participate directly in
the collisional processes that establish thermal equilibrium within
the phonon distribution. Hence it is natural to inquire as to how
effectively a mode that is cooled irreversibly can equilibrate with the
reservoir of other phonons in the solid. Restoration of equilibrium
in solids is governed by Umklapp processes, phonon collisions
requiring a reciprocal lattice vector G to conserve momentum
[33]. Zone center phonons have low wavevectors, so when two
phonons interact with a third via an Umklapp process, there is no
way to satisfy the relation k1 + k2 = k3 + G with a wavevector G that
spans the entire Brillouin zone. To achieve uniform cooling of even
a single mode, phonons propagating along orthogonal axes should
be addressed, and this calls for the introduction of two more sets of
counterpropagating Raman beam pairs to cover all three orthogonal
space axes. Additionally, beam switching along each axis is needed
to interrogate both forward- and backward-traveling phonons of the
same frequency. Fast directional switching of each beam pair can

be implemented with a Pockel’s cell that reverses the propagation
direction of the ω1 and ω2 beams simultaneously within the sample
[28] on a timescale much shorter than the lifetime of the 2F7/2
shelving state.
While uniform cooling of one mode can be assured by such
procedures, thermal equilibration of the sample as a whole relies on
Umklapp processes that involve phonons of wavevector G . Although
the occupation probability of such phonons drops exponentially
with decreasing temperature, the thermal conductivity that is
proportional to the phonon mean free path actually rises through
most of the cryogenic range. Hence sample equilibration times
drop until temperatures around 10–20 K are reached. Then the
mean free path becomes limited by sample dimensions and thermal
conductivity drops as T3 due to its proportionality to specific heat
[33]. The consequence of this is that fast thermal equilibration can
be anticipated in Raman laser cooling of solids at all but the very
lowest (T < 10 K) temperatures.
References
1. Pringsheim, P. (1929). Zwei Bemerkungenüber über den Unterschied

von Lumineszenz- und Temperaturstrahlung, Z. Phys. A, 57, pp. 739–
746.
2. Hansch, T. W., Schawlow, A. L. (1975). Cooling of gases by laser
radiation, Opt. Commun., 13, pp. 68–69.
3. Landau, L. (1946). On the thermodynamics of photoluminescence, J.
Phys. (Moscow), 10, pp. 503–506.
4. Ruan, X. L., Rand, S. C., Kaviany, M. (2007). Entropy and efficiency in
laser cooling of solids, Phys. Rev. B, 75, pp. 214304(1–9).
5. Ashkin, A., Dziedzic, J. M. (1987). Optical trapping and manipulation of
viruses and bacteria, Science, 235, pp. 1517–1520.
6. Cohadon, P. F., Heidmann, A., Pinard, M. (1999). Cooling of a mirror by
radiation pressure, Phys. Rev. Lett., 83, pp. 3174–3177.
7. Metcalf, H., van der Straten, P. (1999). Laser Cooling and Trapping
(Springer-Verlag, New York).
Nature, 377, pp. 500–503.
References 209
9. Powell, R. C. (1998). Physics of Solid State Laser Materials (Springer-

Verlag, New York), pp. 154.
10. Loudon, R. (2000). The Quantum Theory of Light, 3rd Ed. (Oxford
University Press, Oxford), p. 349.
11. Donovan, B., Angress, J. F. (1971). Lattice Vibrations (Chapman and
Hall, London).
12. Clark, J. L., Miller, P. F., Rumbles, G. (1998). Red edge photophysics of
ethanolic rhodamine 101 and the observation of laser cooling in the
condensed phase, J. Phys. Chem. A, 102, pp. 4428–4437.
semiconductor by 40 Kelvin, Nature, 493, pp. 504–508.
14. Bahl, G., Tomes, M., Marquardt, F., Carmon, T. (2012). Observation of
spontaneous Brillouin cooling, Nat. Phys., 8, pp. 203–207.
15. Kittel, C. (1971). Introduction to Solid State Physics, 4th Ed. (John Wiley
& Sons, New York).
17. Pines, D. (1964). Elementary Excitations in Solids (W.A. Benjamin, Inc.,
New York), p. 232.
18. Shen, Y. R. (1984). The Principles of Nonlinear Optics (John Wiley &
Sons, New York), p. 192.
19. Tomes, M., Marquardt, F., Bahl, G., Carmon, T. (2011). Quantum
mechanical theory of optomechanical Brillouin cooling, Phys. Rev. A,
84, pp. 063806(1–5).
20. Shen, Y. R., Bloembergen, N. (1965). Theory of stimulated Brillouin and
Raman scattering, Phys. Rev., 137(6A), pp. 1787–1805.
21. Yariv, A. (1965). Quantum theory for parametric interactions of light
and hypersound, IEEE J. Quantum Electron., 1(1), pp. 28–36.
22. Louisell, W. H. (1960). Coupled Modes and Parametric Electronics
(Wiley, New York).
23. Toschek, P. E. (1985). Lumino-refrigeration, Ann. Phys., 10, pp. 761–
775.
24. Lindberg, M., Javanainen, J. (1986). Temperature of a laser-cooled
trapped three-level ion, J. Opt. Soc. Am. B, 3(7), pp. 1008–1017.
25. Kasevich, M., Chu, S. (1992). Laser cooling below a photon recoil with
three-level atoms, Phys. Rev. Lett., 69, pp. 1741–1744.
26. Rand, S. C. (2013). Raman laser cooling of solids, J. Lumin., 133, pp.
10–14.
27. Reichel, J., Morice, O., Tino, G. M., Salomon, C. (1994). Subrecoil Raman
cooling of Cesium atoms, Europhys. Lett., 28, p. 477.
28. Reichel, J., Bardou, F., Ben Dahan, M., Peik, E., Rand, S. C., Salomon, C.,
Cohen-Tannoudji, C. (1995). Raman cooling of Cesium below 3 nK:
new approach inspired by Levy flight statistics, Phys. Rev. Lett., 75, pp.
4575–4578.
29. Rand, S. C. (2015). Lectures on Light, 2nd Ed. (Oxford University Press,
Oxford), Appendix G.
30. Kuhn, A., Perrin, H., Hansel, W., Salomon, C. (1997). Three dimensional
Raman cooling using velocity selective rapid adiabatic passage, in OSA
TOPS on Ultracold Atoms and BEC, 7, pp. 58–66.
31. Feofilov, S. P., Kulinkin, A. B., Gacoin, T., Mialon, G., Dantelle, G., Meltzer,
R. S., Dujardin, C. (2012). Mechanisms for Ce3+ excitation at energies
below the zero-phonon line in YAG crystals and nanocrystals, J. Lumin.,
132, pp. 3082–3088.
32. Abragam, A., Bleaney, B. (1970). Electron Paramagnetic Resonance of
Transition Ions (Dover, New York), pp. 560–562.
33. Kittel, C. (1971). Introduction to Solid State Physics, 4th Ed. (Wiley, New
York), p. 228.
Chapter 7
Deep Laser Cooling of Rare Earth–Doped

Crystals by Stimulated Raman Adiabatic
Passage
Andrei Ivanov, Yuriy Rozhdestvensky, and Evgeniy Perlin

Center of Information Optical Technologies, ITMO University,
Birzhevayaliniya, 14, Saint Petersburg, 199034, Russia
avivanov@mail.ifmo.ru
7.1 Introduction
Cooling of solid-state matter by laser radiation has definite
advantages over other types of cooling such as thermoelectric,
mechanical, and fluid. Optical coolers are compact, they have no
vibrations, and they use light as a coolant. Such coolers can be
effectively utilized in the case of extremely small solid-state objects
that are used in up-to-date optoelectronics. In the list of possible
applications of laser cooling in solids there are self-cooled lasers,
thermostabilized light-emitting diodes (LEDs), and space sensors.
Although different solid-state materials were proposed for laser
cooling in the last few decades, great success has been achieved to

www.panstanford.com
212 Deep Laser Cooling of Rare Earth–Doped Crystals by Stimulated Raman Adiabatic Passage
date for glasses and crystals doped by the rare earth (RE) ions. There
are two reasons that ensure the success of laser cooling of such
materials. The first reason is a strong coupling between electrons
and vibrations in the RE ion. The ion electronic structure consists
of levels that are split into sublevels by a crystal field of the matrix
due to the Stark effect. The energy gaps between Stark-split sublevels
(SSSs) are comparable with phonon energies of the crystal matrix.
So, there are conditions to observe the anti-Stokes fluorescence
(ASF) in these solid systems. The second reason is the development
of technology that allows producing doped materials of high purity.
It means that in such RE ion systems there are low concentrations
of unwanted impurities. Reducing the impurity concentrations
decreases background absorption and unintended heating by a laser
radiation.
Laser cooling of a solid was originally attained in a Yb3+-doped
fluorozirconate glass that was cooled down by 0.3°С below room
temperature in 1995 [1]. Since that time, laser cooling was observed
in various glasses and crystals doped by the RE ions [2–7]. At present,
progress of the technology makes it possible to reach the range of
cryogenic temperatures for a Yb:YLF crystal [8]. This result is better
than the results obtained on cooling with the use of thermoelectric
Peltier elements.
The traditional model that describes the optical refrigeration of
doped solids is based on a four-level system of 4f multiplet states
of an RE ion [6]. In this model, the ASF is provided by (i) resonance
transitions between the upper ground and lower excited levels, (ii)
phonon absorption between the SSSs, and (iii) subsequent radiative
relaxation to the ground states. At the same time, in such a model,
the rate of the cooling cycle is very low because of the small cross
section of forbidden transitions between the levels of the same ion
configuration and because of the low population of the upper ground
level. For improving the cooling performance in the traditional
model, a nonresonant cavity for maximizing the absorbed laser
power is used [6].
In the last few decades, the coherent phenomena stimulated
by laser radiation, such as coherent population trapping (CPT) and
electromagnetically induced transparency (EIT), have been studied
in atomic physics most intensively [9]. At the same time, the CPT and
EIT effects, including the room-temperature case, were demonstrated
Introduction 213
for solid systems by several groups [10–12]. For the CPT effect to
be observed, it is very important to comply with certain relations
between the lifetimes of the electron levels involved in the cooling
process. For example, for a Λ system, in which there exist three levels
(the ground and first and upper excited levels), the presence of one
metastable level with a long lifetime (the first excited level) provides
optical interference phenomena. Such a situation is possible in RE-
doped crystals due to the existence of dipole-allowed 4f–5d and
forbidden 4f–4f optical transitions.
To improve the characteristics of the cooling process, such as
cooling efficiency, a minimum attainable temperature, and a cooling
rate, Raman mechanisms were recently proposed [13–15]. Raman
cooling used laser pulse pumping through dipole-allowed 5d ion
levels by the two-photon Raman (TPR) scattering proposed for
doped materials (see Section 6.4). The mechanism is based on 4f–5d
and 5d–4f optical transitions involving longitudinal optical phonons
of the crystal matrix. The acceleration of the cooling cycle is achieved
by combining the stimulated TPR process with third pulse pumping
at the allowed fluorescence transition. Nevertheless, in this model,
the cooling efficiency is 0.6%. This is because of the low transition
rate of the two-photon process that involves phonons. Moreover,
this cooling mechanism stops working after freezing the vibrations
of the optical modes. Another Raman mechanism of cooling is based
on methods of coherent pumping of doped crystals [14, 15]. The
coherent and complete population transfer between the ground
and the first excited levels of 4f multiplet through the 5d ion levels
was achieved by using the different Raman techniques, namely TPR
scattering, stimulated Raman adiabatic passage (STIRAP), and the
π-pulse method. It was shown in these works that the coherent
pumping increases the number of electrons that participate in the
cooling cycle, which leads to increasing of the cooling power. At
the same time, the results showed that the deformation potential
mechanism for description of the interaction between ion
electrons and vibrations of the crystal medium failed to explain the
establishment of the Boltzmann distribution of the electrons at the
SSSs.
In this chapter, we consider a cooling model with STIRAP for
optical pumping of an RE-ion-doped crystal, taking into account
vibronic interaction at the SSSs of ion levels. In the model, cooling
and heating processes, which make substantial contributions to the

cooling mechanism, are defined. The cooling processes are attributed
to the vibronic coupling of the SSSs. In addition, using vibronic
interaction allows one to achieve the Boltzmann distribution of the
electrons at the SSSs in a framework of the proposed model. We apply
density matrix formalism for determining the time dependence
of electron populations at the ion levels under the action of pulse
pumping. Then, the level populations are used for estimation of
cooling characteristics of the model. The calculations, which are
obtained for the Yb3+:CaF2 system, show that coherent pumping
by STIRAP leads to deep laser cooling of doped crystals that have
technologically attainable purity to date.
7.2 Vibronic Model of Laser Cooling of Rare

Earth Ions
Let us consider a crystal doped by trivalent RE ions. The specific
feature of the system under consideration is the existence of 5d ion
levels below the conduction band edge of the crystal. In addition,
there are only two 4f ion levels (the ground and first excited levels)
below the 5d levels. We assume that we are dealing with a five-level
energy structure (see Fig. 7.1), which consists of one level of the 5d
configuration (denoted by |3Ò) and four levels of the 4f configuration:
one ground level (denoted by |1Ò) and three SSSs of the first excited
level (denoted by |2Ò,|4Ò, and |5Ò, respectively). In the model, we
assume that the 5d ion level is empty. The energy difference between
the ground level |1Ò and the lower SSS |2Ò is assumed to be large
enough for radiative relaxation from the first excited level to occur
and, thus, to prevent the system from significant heating [16].
We assume that the laser cooling cycle of the five-level system
contains the following steps: (i) optical pumping of the SSS |2Ò,
(ii) interaction of the electrons with the ion vibrations, and (iii)
transference of the population to the ground level |1Ò by radiative
decay (Fig. 7.1). We denote this cooling process as C1. For laser
cooling of the considered system, we use a sequence of two laser
pulses (Fig. 7.2b) with electromagnetic field frequencies ω1 and ω2,
respectively. The pulse sequence is used to implement the STIRAP
in which laser pulses are partially overlapped and come in the
Vibronic Model of Laser Cooling of Rare Earth Ions 215
counterintuitive order when the first pulse initially couples the two
empty levels, |3Ò and |2Ò, and the second pulse couples levels |1Ò
and |3Ò. Such coherent pumping gives rise to the transfer of all the
population from level |1Ò to level |2Ò without populating level |3Ò [14].
The establishment of the Boltzmann distribution at levels |2Ò, |4Ò,
and |5Ò is the result of vibronic interaction. This interaction leads to
consumption of vibrational energy of the RE ions and consequently
to the cooling of the system. Finally, the population returns to the
ground level by radiative decay, and the cycle is completed.
Conduction band
|3〉 5d level
Δ1
Δ2
ω1 ω2
ω02 4f levels
|5〉
|4〉
|2〉
ω01
γ
|1〉
i ii iii
Figure 7.1 Scheme of the laser cooling cycle for the five-level energy
structure of an RE ion: (i) two-photon optical pumping, (ii)
vibronic interaction, and (iii) radiative decay. Solid arrows
show the optical transitions at the frequencies ω1 and ω2;
dotted arrows show the vibronic interaction with energies ω01
and ω02; downward dashed arrows refer to the nonradiative
decay; wavy arrows show the radiative decay γ; Δ1 and Δ2
are the detunings of optical transitions; the detunings of the
transitions under the vibronic interactions are not depicted
(see text).
It should be noted that the actual efficiency of the above-

proposed cooling cycle is defined by the lack of population of 4fn–1
5d levels. This is necessary to avoid heating processes. In addition,
the efficiency of the cooling cycle is defined by the creation of a high
population of SSSs to stimulate vibronic interaction. Such conditions
can be achieved on coherent interaction of radiation with the doped

crystal and based on the differences between the energy relaxation
times of the 5d and 4f levels. The 5d levels rapidly decay to lower
levels, with a lifetime of the order of tens of nanoseconds [17],
whereas the lifetime of the 4f configuration levels is of the order of
milliseconds [16]. This relationship between the decay times of the
excited levels provides the conditions for the cooling process.
7.2.1 Vibronic Interaction

Vibronic coupling is a key factor for determining the optical properties
of local centers in a crystal [18, 19]. In case of quasi-degenerate or
weak split electronic states, these properties are governed by the
pseudo Jahn–Teller effect (PJTE) [20]. The small intervals in the
electronic spectrum mean a relatively slow electron movement that
is comparable with a nuclear movement. So, the movements of the
electrons and the nucleus in the RE ion cannot be divorced and a
nonadiabaticity of the coupling must be taken into account. It should
be emphasized that in case of the RE ion with unfilled d or f electron
shells, there is a strong interaction between localized electrons and
local or pseudolocal vibrations [21].
Vibronic interaction exerts a strong influence on the laser cooling
process. The existence of vibronic coupling leads to decreasing of
vibrational energy of the RE ion interacting with laser radiation
under cooling conditions. Since the ion is bound with the atoms of
the crystal by elastic interaction, laser cooling of the RE ion gives
rise to cooling of the crystal matrix as a result of the achievement
of thermal equilibrium. The Boltzmann distribution at SSSs is a
sequence of the coupling between the vibronic subsystem and the
phonon reservoir.
It is worth mentioning that the use of deformation potential
formalism to describe the interaction of ion electrons with
lattice phonons fails to explain the establishment of a Boltzmann
distribution at SSSs of the excited RE ion level [15].
7.2.2 Model Hamiltonian for the Rare Earth Ion System

For a theoretical description of the time evolution of the five-
level system of the RE ion, we use strict quantum mechanical
treatment that is based on density matrix formalism. We consider

the quantum mechanical system that consists of electronic and
local, or pseudolocal, vibration subsystems. We assume that the
energy spectra of the noninteracting subsystems are known. So, the
Hamiltonian of the system without interaction can be written in the
following form:
H0 = H0e + H0 v, H0e = Âe a a , H
i
+
i i i 0v = Â w
n
0n bn bn
+
(7.1)
Here H0e is the Hamiltonian of the electronic subsystem and H0v is

the Hamiltonian of the pseudolocal mode; ai+ and ai are the creation
and annihilation operators of an electron at the ion level i of energy εi;
bn+ and bn are the creation and annihilation operators of a pseudolocal
mode n of frequency ω0n. The subsystems interact with each other
and with linearly polarized electromagnetic waves. We assume that
there is a strong linear vibronic coupling [22, 23] and there are two
Jahn–Teller active modes of frequencies ω01 and ω02 in the system
(Fig. 7.1). The interaction Hamiltonian is taken in the following form:
+
13,l k a+ a exp(i D t ) + a+ a W
Hint = W 31 ,l1k 1 exp( -i D 1t )
1 1 3 1 1 1 3
+
23,l k a+ a exp(i D t ) + a+ a W
+ W 32 ,l2k 2 exp( -i D 2t )
2 2 3 2 2 2 3

+ Q + + +
24 ,01a4 a2exp(id 01t ) + a2 a4 Q 42,01exp( -id 01t )

+ Q + + +
45,02a5 a4exp(id 02t ) + a4 a5Q54 ,02exp( -id 02t ) , (7.2)
ij ,l k = mij E l k , Q
W kij
ij ,0n = b , (7.3)
n
m m m m

Here μij are the matrix elements of the dipole moment operator
+
for levels i and j; E lmk m ( E l k ) is the electric field operator that
m m
contains the annihilation (creation) operator of a photon; λm, and

κm are the polarization and the wavevector of the photon related to
the laser pulse m; Δ1 = ω31 – ω1 and Δ2 = ω32 – ω2 are the detunings
of optical transitions, where ωji are the frequencies of optical
transitions between the levels i and j; kij are the matrix elements of
vibronic coupling for levels i and j; δ01 = ω42 – ω01 and δ02 = ω54 –
ω02 are the detunings of transitions under the vibronic interaction;
and an operator A+ denotes a Hermitian conjugate operator of A.
The coupling of the system with the phonon continuum is taken

into account by the introduction of the relaxation constants in a
framework of density matrix formalism [23].
7.2.3 Time Evolution of the Five-Level System

To estimate the cooling performance, we use a density matrix method
to take into account the coherent interaction of the pulse sequence
with the considered system. Since optical pumping transfers the
population of the ground level |1Ò to the lowest excited level |2Ò,
cooling energy is determined by a number of the electrons moved
to the upper levels |4Ò and |5Ò by the vibronic interaction. Hence, our
goal is to calculate the populations of levels |4Ò and |5Ò using diagonal
elements of the density matrix. To achieve this goal we consider the
dynamics of the system that is described by a set of equations for the
density matrix elements.
We theoretically treat the time evolution of a five-level
system using the interaction representation in the rotating-wave
approximation. In this case, the set of equations is
+
31,l k - i W
r 11 = i r13 W 13,l k r + g r + g r + g r + g r ,
1 1 1 1 31 22 22 33 33 44 44 55 55
13,l k r + i r W
r 12 = i ( D1 - D 2 )r12 - i W + +
1 1 32 13 32,l2k 2 + i r14 Q 42,01 - g 12r12 ,
13,l k - i W
r 13 = i r11 W 13,l k r + i r W
1 1 1 1 33 12 23,l2k 2 + i D 1 r13 - g 13r13 ,
13,l k r + i r Q
r 14 = i (d 01 - D 2 + D1 )r14 - i W +
1 1 34 12 24 ,01 + i r15 Q54 ,02 - g 14 r14 ,
13,l k r + i r Q
r 15 = i (d 01 + d 02 - D 2 + D1 )r15 - i W
1 1 35 14 45,02 - g 15r15 ,
+
32
r 22 = i r23 W +
,l2k 2 - i W23,l2k 2 r32 + i r24 Q 42,01 - iQ 24 ,01 r42
+g 44
nr
r44 + g 33 r33 - g 22 r22 ,
23,l k - i W
r 23 = i D 2r23 + i r22 W 23,l k r - iQ

2 2 2 2 33 24 ,01 r43 + i r21 W13,l1k 1 - g 23r23 ,
+

r 24 = i d 01r24 + i r22 Q
24 ,01 - iQ 24 ,01 r44 - i W23,l2k 2 r34 + i r25 Q54 ,02 - g 24 r24 ,
nr

r 25 = i (d 01 + d 02 )r25 - iQ
24 ,01 r45 - i W23,l2k 2 r35 + i r24 Q 45,02 - g 25 r25 ,
nr
+
13,l k - i W
r 33 = i r31 W 31 +
,l1k 1 r13 + i r32 W23,l2k 2 - i W32,l2k 2 r23 - 4g 33r33 ,
1 1
+
31
r 34 = i (d 01 - D 2 )r34 - i W +
,l1k 1 r14 - i W32,l2k 2 r24
+

+i r32 Q
24 ,01 + i r35 Q54 ,02 - g 34 r34 ,
+
31
r 35 = i (d 01 + d 02 - D 2 )r35 - i W +
,l1k 1 r15 - i W32,l2k 2 r25 + i r34 Q 45,02 - g 35r35 ,

r 44 = i r42 Q + +
24,01 - iQ 42,01 r24 + i r45 Q54 ,02 - iQ 45,02r54
-(g 44 + g 44
nr
)r44 + g 55
nr
r55 + g 33 r33 ,
+ r + ir Q
r 45 = i d 02 r45 - iQ
42,01 25 44 45,02 - iQ 45,02 r55 - g 45 r45 ,

r55 = i r54 Q +
45,02 - iQ54 ,02 r45 - (g 55 + g 55 )r55 + g 33 r33. (7.4)
nr
Other matrix elements are determined by the relation

ρij = ρ*ji. In addition, the particle number conservation condition
ρ11 + ρ22 + ρ33 + ρ44 + ρ55 = 1 must be met. In Eq. 7.4, the quantities
gij describe radiative relaxation of the density matrix elements, and
gijnr describe the nonradiative relaxation. Equation 7.4 is written not
taking into account nonradiative relaxation from the SSSs because
of the short duration of the laser cooling cycle compared to the
relaxation time. This set of equations can be solved only numerically.
In our calculations, we determine the expectation values of the
level populations by statistical averaging over states of the photonic
and vibrational subsystems. The averaging procedure is based
on the standard statistical approximation in which the photonic
and vibrational subsystems can be described by the equilibrium
distribution functions. This means that the density matrix of the
complete system is replaced by the product of the density matrices
of the subsystems
ρ(t) = ρe(t)ρph(0)ρv(0), (7.5)
where subscripts e, ph, and v denote the electronic, photonic,
and vibrational subsystems, respectively. In case of a photonic
subsystem, we use the Glauber–Sudarshan P-representation with a
δ-like diagonal weight function for single-mode laser radiation for
each pulse [24]:

Ú
rph (0) = J ({xk }) ’xk
k xk d 2xk ,
1
J ( xk ) = d ( xk - nk ) (7.6)
2p nk
Here xk and xk are the eigenvalues and eigenstates of the
annihilation or the creation operator; nκ is the average photon
number in the mode. And the vibrational subsystem is described by
the density matrix operator
Ê ˆ
rv (0) = Z -1 exp Á -
ÁË Â b w
j
T
+
0 jbj bj ˜
˜¯
,
È Ê ˆ˘

Í ÁË j Â
Z = Tr Íexp Á - bT w0 j b+j b j ˜ ˙ , bT -1 = kBT .
˜¯ ˙
(7.7)
Î ˚
where kB is the Boltzmann constant and T is the temperature of
the system. In the case of a pseudolocal vibrational subsystem,
the averaging over the states is performed using the local density
of states (DOS) that is approximated by the δ-function for each
frequency ω01 and ω02 in our calculations. We assume the resonance
vibronic interaction that allows us to use the detunings δ01 = 0 and
δ02 = 0. Thus, taking into account the averaging procedure, we can
change the boson annihilation and creation operators by c-numbers
n exp(j ), where n is the average particle number in the mode and
φ is the phase of the eigenvalue of the annihilation or the creation
operator.
The time dependence of the diagonal elements of the reduced
density matrix for the first cooling cycle is shown in Fig. 7.2a. For
implementation of laser cooling we use two identical pulses that
have a Gaussian shape:
Ê (t - t0 )2 ˆ
E0 (t ) = E0 exp Á - ˜ . (7.8)
Ë s2 ¯
The width of each pulse is σ = 2.0 × 10–7 s and the electric field
amplitude of each pulse is E0 = 1.2 × 103 V/cm. The distance between
the pulses is 9.8 × 10–8 s (Fig. 7.2b). The first pulse couples the two
empty levels |3Ò and |2Ò, and the second pulse couples levels |1Ò and
|3Ò (see Fig. 7.1). The shapes and parameters of the pulses are chosen
to satisfy the adiabatic conditions: (i) a smooth pulse, (ii) a long
interaction time, and (iii) a large Rabi frequency [25]. Hence, the
adiabatic passage process can be actualized if the Rabi frequencies

are sufficiently larger than the inverse time of the pulse overlap.
1.0
0.8
0.6
0.4
0.2
0.0
–7 –7
0.0 2.0×10 4.0×10
–7 –7
0.0 2.0×10 4.0×10
Figure 7.2 (a) Time dependencies of the diagonal elements of the density
matrix: r11 (solid line), r22 (dashed line), r44 (dotted line), and
r55 (dash-dotted line). (b) Rabi frequencies of laser beams.
Note that the sequence of pulses is counterintuitive: first is
W32 (solid line) and second is W13 (dashed line). Results are
obtained at 300 K (cooling process C1).
The time dependencies of the diagonal elements of the density

matrix are shown in Fig. 7.2a. It is shown that the total population
of level |1Ò is transferred to levels |2Ò, |4Ò, and |5Ò throughout the
STIRAP without populating level |3Ò. In the calculations, we use the
detunings D1 = 0 and D2 = 0. This method of pumping can also be
used if the detunings D1 and D2 are not equal to zero. In the case of
substantial detunings, the total population of the ground level can be
transferred if the electric field amplitude will be increased.
In the calculations, we use the parameters typical of the solid
systems under study: mij = 3.34 × 10–28 C cm; ħω01 = 191 cm–1; ħω02
= 487 cm–1; g33 = 1.0 × 108 s–1; g13 = g23 = g34 = g35 = 4g33; g = 1.0 × 103
s–1; g12 = g14 = 0.5 g; g22 = g44 = g; gnr = 1.0 × 1012 s–1; g44nr = g55nr = gnr;
g24nr = g45nr = 0.5 gnr; g25nr = 0; and cycle duration τ = 4 × 10–7 s. The
lattice temperature was assumed to be T = 300 K for the first cooling
cycle. The expectation values of quantities Qij were determined to
satisfy the condition of establishment of the Boltzmann distribution
at the levels |2Ò, |4Ò, and |5Ò at 300 K. Thus, the strengths of vibronic
interaction were assumed to be k24/ħ = 5.1 × 1011 s–1 and k45/ħ =
4 × 1011 s–1 in the calculations (see Eq. 7.3). The frequencies of the
laser beams and optical transitions for certain system are discussed
below.
7.3 Laser Cooling of the Yb3+:CaF2 System

For the real system that can be described by the model of laser
cooling developed in Section 7.2, we can choose the calcium fluoride
crystal doped with trivalent ytterbium, Yb3+:CaF2. This material has
the following features that are important in considered case. The
lowest 5d level of the Yb3+ ion falls within the bandgap of CaF2. The
4f multiplet of the Yb3+ ion involves only two levels, 2F7/2 (ground
level) and 2F5/2 (excited level), that are split into four (E1, E2, E3, E4)
and three (E5, E6, E7) SSSs, respectively. The energy gap between
the ground and excited levels is 9523 cm–1. Such a large bandgap
allows for not taking into account the nonradiative relaxation to the
ground level. The energy diagram of the 4f multiplet is shown in
Fig. 7.3 [26]. The optical spectrum of the CaF2 crystal doped by Yb3+
ions manifests a rich vibronic structure [27]. For example, detailed
analysis of the phonon side band shows the local breathing-mode
Laser Cooling of the Yb3+:CaF2 System 223
oscillations of frequency 450 cm–1 that is close to the frequencies

of the electronic transitions ω23 and ω67 of the Yb3+ ion (see Fig.
7.3). It is noteworthy that for this specific system, we use the same
parameters as those listed in Section 7.2.
5d 68741
ω1 ω2
2
F5/2
E7 10887
E6 10400
E 10209
2
F7/2
E4 686
E3 531
E2 110
E1 Yb 3+
0
Figure 7.3 Energy diagram of the Yb3+ ion and the laser radiation photon
energies used in calculations (in cm–1). Downward dashed
arrows denote the processes of radiative (nonradiative)
relaxation from the excited to ground sublevels. Only one set of
relaxation processes from level 7 is depicted (see text).
To implement coherent pumping of the doped system chosen

here, vacuum ultraviolet (VUV) lasers must be used [28]. We assume
that the photon energy of the radiation for the first laser pulse, ω1,
is 68,055 cm–1 and falls in resonance with the electron transition
between the highest ground SSS (denoted by E4) and the lowest 5d
level of the Yb3+ ion. The photon energy for the second laser pulse,
ω2, is 58,532 cm–1 and falls in resonance with the transition between
the lowest 5d level and the lowest SSS (denoted by E5) of the excited
level (see Fig. 7.3). The relation between photon energies is chosen
to avoid the population of the 4f 125d levels by optical transitions
from the ground SSSs and, so hence, to avoid additional heating of
the system.
7.3.1 Number of Electrons Involved in the Cooling Cycle

The efficiency of the proposed cooling model depends on the
number of electrons involved in the cooling cycle. For determining
the number of electrons that are transferred to the excited level from
the ground level, we use a simplified five-level model that takes into
account vibronic interaction at the SSSs the both RE ion levels (Fig.
7.4). For this goal, we modify Eq. 7.4 in accordance with the scheme
depicted in Fig. 7.4 to a new set of equations:
+
r 11 = i r13 Q
31 ,l1k 1 - iQ13,l1k 1 r31 + g 33r33 + g 44 r44 + g 55r55 ,

r 12 = -iQ +
13,l1k 1 r32 + i r13 Q32,l2k 2 + i r14 W 42,01 - g 12 r12 ,
nr

r 13 = i r11 Q +
13,l1k 1 - iQ13,l1k 1 r33 + i r12 Q23,l2k 2 - g 13 r13 ,
nr

r 14 = -iQ +
13,l1k 1 r34 + i r12W24 ,01 + i r15 Q54 ,02 - g 14 r14 ,

r 15 = -iQ
13,l1k 1 r35 + i r14 Q 45,02 - g 15r15 ,

r 22 = i r23 Q + +
23,l2k 2 - iQ23,l2k 2 r32 + i r24 W 42,01 - i W24 ,01 r42
+g 55 r55 + g 44 r44 - g 22

nr
r22 ,
+
r 23 = i r22 Q + ,
23,l2k 2 - iQ 23,l2k 2 r33 - i W24 ,01 r43 + i r21 Q13,l1k 1 - g 23 r23
nr
24 ,01 - i W
r 24 = i r22 W +
24 ,01r - iQ +
44 23,l2k 2 r34 + i r25 Q54 ,02 - g 24 r24 ,
r 25 = - i W +
24 ,01r - iQ
45 23,l2k 2 r35 + i r24 Q 45,02 - g 25r25 ,

r 33 = i r31 Q + +
13,l1k 1 - iQ31 ,l1k 1 r13 + i r32 Q23,l2k 2 - iQ32,l2k 2 r23
-g 33
nr
r33 + g 22
nr
r22 + g 44 r44 + g 55 r55,
+
r 34 = -iQ +
31 ,l1k 1 r14 - iQ32,l2k 2 r24 + i r32 W24 ,01 + i r35 Q54 ,02 - g 34 r34 ,
+
r 35 = -iQ
31 ,l1k 1 r15 - iQ32,l2k 2 r25 + i r34 Q 45,02 - g 35r35 ,
+
24 ,01 - i W
r 44 = i r42 W 42 +
,01 r24 + i r45 Q54 ,02 - iQ 45,02 r54 - 3g 44 r44 + g 55 r55 ,
nr
+
42
r 45 = -i W
,01 r25 + i r44 Q 45,02 - iQ 45,02 r55 - g 45 r45,
nr

r55 = i r54 Q +
45,02 - iQ54 ,02 r45 - 3g 55 r55 + g 55 r55 . (7.9)
nr
In the equations, we assume that all of the detunings are zero. In

Eq. 7.9 we save the notations used in Eq. 7.4.
|5Ò
ω01
|4Ò
|2Ò
ω04
|3Ò
ω03
|1Ò
Figure 7.4 Scheme of vibronic interaction at the SSSs of the ground and
excited RE ion levels coupled by the electromagnetic field for
calculation of the number of electrons involved in the cooling
process. The five-level energy structure is used for simplicity
of calculations. The solid arrow shows the optical transitions
at the frequencies ω; dotted arrows show the vibronic
interaction with frequencies ω01, ω03, and ω04; and downward
dashed arrows refer to the nonradiative decay.
The changing of the level populations at the SSSs under the action
of the Gaussian laser pulse is depicted in Fig. 7.5. It is shown that
electrons from all SSSs of the ground level participate in the cooling
cycle at 300 K. Moreover, the aggregate population that is transferred
to the excited levels |4Ò and |5Ò after the pulse action is larger than
the population at the upper level |2Ò before the pulse action (Fig.
7.4). Thus, the existence of vibronic interaction at SSSs leads to an
increase of the number of electrons involved in the cooling cycle.
Since the electrons initially populated the ground levels |1Ò and |3Ò
consume energy to transfer to the upper level |2Ò, these electrons
participate in another cooling process that is additional to the
process described in Section 7.2. We denote this process as C2.
In the calculations, we use the following parameters: ħω03 = 421
cm–1; ħω04 = 155 cm–1; g14 = g15 = g24 = g25 = g34 = g35 = g/3; g55 = g44=
g; g22nr = g33nr = g55nr = gnr; g12nr = g13nr = g23nr = g45nr = 0.5 gnr; and T
= 300 K. The Boltzmann distribution at the levels is established if

the strengths of vibronic interaction are k12/ħ = 3.3 × 1011 s–1, k23/ħ
= 4.3 × 1011 s–1, and k45/ħ = 5 × 1011 s–1. Other parameters are the
same as in the Section 7.2.
1.00
0.95
0.90
0.85
0.80
–7 –7 –7
0.0 1.0×10 2.0×10 3.0×10
0.040
0.035
0.030
0.025
0.020
–7 –7 –7
0.0 1.0×10 2.0×10 3.0×10
0.08
0.06
0.04
0.02
0.00
–7 –7 –7
0.0 1.0×10 2.0×10 3.0×10
–7 –7 –7
0.0 1.0×10 2.0×10 3.0×10
Figure 7.5 Time dependencies of the diagonal elements of the density

matrix: (a) r11, (b) r33 (solid line) and r22 (dashed line), and
(c) r44 (solid line) and r55 (dashed line). (d) Rabi frequency of
laser beam. Results are obtained at 300 K (cooling process C2).
7.3.2 Estimation of Cooling Performance

For determining cooling performance, we must take into account
all the processes that contribute to cooling and heating of the
considered doped system. The total power left over in the system
is determined by the difference between heating (Pheat) and cooling
(Pcool) powers:
Ptot = Pheat – Pcool, Pheat = Pim + Pvib. (7.10)
Here, Pim is the power absorbed by impurity ions and Pvib is the
power released in consequence of electronic transitions that lead
to the energy transfer to the vibrational subsystem at the SSSs.
The cooling power is determined by two above-considered cooling
processes that are denoted by C1 and C2, respectively. Since the
average thermal energy per one electron e vib
C
is removed from the
host matrix by radiative decay from the excited level with the rate g,
the expression for the cooling power can be written as
(
Pcool = N1g e vib
C1
+ e vib
C2
, ) (7.11)
S Ê Sˆ Ê Sˆ
e vib
C1
= r22
1
(t ) + r44
1
(t ) Á E6 - E5 + ˜ + r44
1
(t ) Á E7 - E5 + ˜ ,
4 Ë 4 ¯ Ë 4¯
(7.12)
e vib
C2
(
= r11
2
(0) - r11
2
(t ) ) (E 4 (
- E1 ) + r33
2
(0) - r33
2
(t ) ) (E 4 - E3 ) ,
(7.13)
( )
3
S ∫ 3E 4 - ÂE
m =1
m, N1 = nion Ne r44
2
(t ) + r55
2
(t ) . (7.14)
Here riin is the averaged value of the population of the level

i for the system of differential equations n(n = 1 denotes Eq. 7.4; n
= 2 denotes Eq. 7.9); N1 is the number of electrons involved in the
cooling cycle; nion is the ion concentration in the crystal; and Ne is
the number of electrons at the ground level. In Eq. 7.12, we assume
that the rates of radiative decay from all of the excited SSSs to all the
ground SSSs are equal. One set of relaxation processes from level
7 is depicted in Fig. 7.3. The relaxation is related to two heating
processes that we take into account in our calculations. Since the
electron has an equal probability to appear at the ground SSSs as a
result of the relaxation processes, establishment of the Boltzmann
distribution at these SSSs heats the system. We denote this process as

H1. Another heating process is nonradiative decay from the excited
SSSs. The rate of radiative decay is of 4 orders of magnitude higher
than the rate of nonradiative decay [16]. Due to this fact, no account
is taken of nonradiative relaxation in Eqs. 7.4 and 7.9. However, the
contribution of the nonradiative relaxation in heating power must
be included. We denote this process as H2. So, the total contribution
to the heating power can be written as
(
Pvib = N1 ge vib
H1 H2
+ Wnr e vib ) (7.15)
Here Wnr is the nonradiative decay rate. The average thermal

energy per electron that is absorbed in the system can be determined
for heating process H1 as
S1
e vib
H1
= - r11
2
(0) (E2 - E1 ) - r22
2
(0) (E 4 + E3 - E1 ) -
4
- r33

2
(0) (E3 - E1 ) (7.16)
and for heating process H2 as
S2 ÊS ˆ
e vib
H2
= r22
1
(t ) + r44
1
(t ) Á 2 - E6 + E5 ˜
4 Ë 4 ¯
ÊS ˆ
+ r44
1
(t ) Á 2 - E7 + E5 ˜ , (7.17)
Ë 4 ¯
4 3
S1 ∫ Â
m=2
Em - 3E1, S2 ∫ 4 (E5 - E 4 ) - ÂE
m =1
m . (7.18)
The cooling efficiency is defined as a ratio of the negative total

power to the power absorbed in the sample:
Ptot . (7.19)
hcool = -
Pabs
The absorbed power can be taken in the form
Pabs = N1ħ (ω1 – ω 2) λ. (7.20)
Finally, we must determine the power absorbed by impurity
ions. For this goal we can write the ratio of the power absorbed by
impurity ions to the power absorbed in the sample in Eq. 7.19 using
the Beer–Lambert law for both cases:
Pim 1 - exp( - b ) a
= , b = im , (7.21)
Pabs 1 - exp( -1) a rad
where αrad is the optical absorption coefficient for the RE ions and
αim is the absorption coefficient for the impurity ions. Equation 7.21
defines the absorption of laser radiation on the length of αrad–1.
It should be noted that the cooling process in the model
proposed in this study is basically more efficient than the process
involving forbidden optical transitions between the levels of the 4f
multiplet in the traditional model. To demonstrate this advantage of
the proposed model, we directly calculate the difference between
the cooling efficiencies in the above-mentioned two models using
the formulas available in Ref. [6]. Since the fundamental distinction
between the models is that pumping of the SSSs is implemented at
different (allowed and forbidden) optical transitions, the difference
between the cooling efficiencies is defined by the difference between
the absorption efficiencies only (all other parameters of the system
being the same). The absorption efficiency can be expressed as [6]
ηabs = (1 + β)–1. (7.22)
Then, we use the expression for the absorption coefficient to
find quantity β. The expression for the absorption coefficient can be
written as
αi = σiNi, (7.23)
where σi is the absorption cross-section for a certain wavelength
and Ni is the concentration of absorbing ions. If we assume that the
absorption cross sections for the RE and impurity ions are equal,
the quantity β is defined as the ratio of the concentrations of the RE
and impurity ions. Typical commercial fluoride starting materials
have transition metal impurities at the parts per million (parts
in 106) level [6]. And we assume that the crystal doped to the 5
mol% Yb content. So hence, the ratio of the absorption coefficients
is 2.0 × 10–5 in the case of allowed optical transitions. The optical
absorption coefficient controlled by dipole-allowed transitions in
the Yb3+ ion is 2 orders of magnitude larger than the absorption
coefficient controlled by dipole-forbidden transitions. Hence, by
determining the absorption efficiencies for the two models by Eqs.
7.22 and 7.23, we find that the cooling efficiency in the proposed
model with coherent pumping is higher by 0.005 (or by 14%) than
the cooling efficiency in the traditional model with pumping at
forbidden transitions (see Fig. 7.7).
Knowing the cooling power, we can determine the time

dependence of the temperature in the model system under study.
The equation describing the temperature evolution with time can be
written as [29]
dT
C m r0V0 = Ptot (T ) + Pload (T ) . (7.24)
dt
Here, Cm is the specific heat of the cooled sample, ρ0 is the
mass density, V0 is the irradiated volume of sample, and Pload is the
environmental heat load that includes the convective, conductive,
and radiative contributions. On the assumption that convective
and conductive loads are negligible, the environmental heat load is
defined by the radiative load given by the expression
Pload(T) = σBAsεs(Tc4 – T4), (7.25)
where σB is the Stefan–Boltzmann constant; As and εs are the
surface area and emissivity of the sample, respectively; and Tc is the
temperature of the chamber surrounding the sample. Equation 7.25
is written for conditions in which the chamber emissivity is equal to
unity and the chamber surface area is much larger than the surface
area of the sample. Thus, taking into account Eqs. 7.19, 7.20, and
7.25, we can rewrite the differential Eq. 7.24 as
dT
C m r0V0 = -hcool (T )N1(T )(w1 - w2 )g + s B As e s (Tc4 - T 4 ) .
dt
(7.26)
Using Eq. 7.26 we can get the time dependence of the temperature
of the irradiated volume of the sample. For that purpose we must
define how the cooling efficiency and the number of electrons,
which participate in the cooling cycle, depend on the temperature.
The numerical calculations of time dependence of temperature we
organize in the following order. Starting at room temperature, we
find the solutions of the two systems of equations (Eqs. 7.4 and
7.9) for the level populations. After that, we can define the cooling
efficiency and the number of electrons for a certain temperature.
With the use of Eq. 7.26 we determine the temperature decrease
throughout the cooling cycle, the duration of which is equal to γ–1.
Then, we repeat the cycle with the new temperature value. The
calculations are terminated if the temperature stops changing.
The results of numerical simulation with the initial condition

T(0) = Tc = 300 K are shown in Figs. 7.6 and 7.7. In the calculations,
we use the following parameters: V0 = 2.12 × 10–4 cm3, Ne = 13,
nion = 3.7 × 1021 cm–3 (for the Yb:CaF2 crystal doped to the 5 mol% Yb
content), and Wnr = 5 s–1. The energies of the SSSs, Em, are indicated
in Fig. 7.3. Other parameters are defined before. The calculations
are performed for different values of quantity β. The case of dipole-
allowed transitions corresponds to β = 2.0 × 10–5, whereas the other
three values of β are close to the case of dipole-forbidden transitions
with the different background absorption coefficient αim. It is shown
in Fig. 7.6 that in the latter case the minimum attainable temperature
substantially depends on the concentration of transition metal
impurities in the crystal. Reducing the concentration of the impurity
ions leads to deeper cooling of the crystal system. For example, if
β changes from 6.0 × 10–3 to 4.0 × 10–3, the minimum attainable
temperature decreases from 101 K to 85 K (see insert in Fig. 7.6).
This situation is typical for the traditional cooling model with direct
pumping at the 4f–4f transitions [6]. The case of β = 2.0 × 10–5
simulates the situation with coherent pumping through the 5d level
of the RE ion proposed here. In this case, the minimum attainable
temperature is 74 K for current purity materials. It should be noted
that the specified minimum attainable temperatures are reached
within 5 s. The temperature dependencies of the cooling efficiency
for the same values of β are shown in Fig. 7.7. It can be seen that the
cooling efficiency of the doped system decreases if the temperature
decreases. From these dependencies, we can find that the cooling
efficiency in the proposed model (β = 2.0 × 10–5) is higher by 14%
than the cooling efficiency in the traditional model (β = 2.0 × 10–3)
if the optical absorption coefficient controlled by dipole-allowed
transitions is 2 orders of magnitude larger than the absorption
coefficient controlled by dipole-forbidden transitions. In other cases,
the difference is much larger.
In addition, the cooling power in the proposed model is higher
than that in the model with direct pumping at the 4f–4f transitions.
As follows from Eqs. 7.19, 7.20, and 7.26, a substantial contribution
to the increase in the cooling power is made by the increase in the
number of electrons N1 involved in the cooling process, apart from
the increase in the cooling efficiency. In the case of coherent pumping,
the population of the SSSs of the ground level, which is transferred
to the SSSs of the excited level, is larger than the population of the
upper SSS of the ground level of the ion. The ratio of the difference
between these populations to the population of the upper SSS of the
ground level is defined as
r44
2
(t ) + r55
2
(t )
DN = - 1 . (7.27)
r22
2
(0)
Figure 7.6 Time dependencies of the temperature in the process of laser

cooling at different β: (solid line) 6.0 × 10–3, (dashed line)
4.0 × 10–3, (dotted line) 2.0 × 10–3, and (dash-dotted line)
2.0 × 10–5.
The temperature dependence of ΔN is shown in Fig. 7.8. From this

dependence, it can be seen that the number of electrons from lower
SSSs of the ground level, participating in the cooling cycle, increases
with decreasing temperature. It is shown in Fig. 7.8 that below 120
K, the difference of the populations in the order of magnitude is
greater than the population of the upper SSS of the ground level.
This effect is due to the presence of coherent pumping and the
existence of vibronic interaction. In the case of direct pumping, this
population is divided between the ground and excited states in half
(in the limiting case of saturation of absorption). Consequently, at
the excitation intensities corresponding to saturation of absorption

at the 4f–4f transitions, the cooling power in the model of coherent
pumping is several times larger than the cooling power in the model
of direct pumping of electron levels of the RE ion.
Figure 7.7 Temperature dependencies of the cooling efficiency at different

β: (solid line) 6.0 × 10–3, (dashed line) 4.0 × 10–3, (dotted line)
2.0 × 10–3, and (dash-dotted line) 2.0 × 10–5.
Figure 7.8 Temperature dependence of the ratio of the difference

between the population of the SSSs of the ground level, which
is transferred to the SSSs of the excited level, to the population
of the upper SSS of the ground level (see Eq. 7.27).
Conclusions 235
7.4 Conclusions
A model of optical cooling of RE-doped crystals on coherent pulse
pumping of the SSSs of RE ions is considered in this chapter. It is
shown that the use of STIRAP through the dipole-allowed 5d
level of the RE ions improves the characteristics of the process of
laser cooling of solid systems in comparison to the corresponding
characteristics obtained on pumping at the forbidden 4f–4f
transitions. The absorption efficiency in the model proposed in the
study is 14% higher than the absorption efficiency in the traditional
model of direct 4f–4f pumping. In addition, in the proposed model,
the power and cooling rate of the system are increased at least by a
factor of 2 compared to these parameters in the traditional model.
This is due to the fact that the number of electrons involved in the
cooling cycle is larger in the case of coherent pumping than in the
case of direct 4f–4f pumping. As a result, it becomes possible to
implement deeper cooling for Yb3+-doped crystals with purity that
can be currently attained. This advantage of the proposed model
seems to be of particular importance. Our calculations show that
the minimum attainable temperature substantially depends on
the concentration of transition metal impurities in the case of the
traditional model. In this context, it is worth noting that the key
method of improving the cooling efficiency in the case of pumping
at forbidden optical transitions is to reduce the content of side
impurities in the crystal matrix, which presents a rather difficult
technological problem. Moreover, the proposed model is not
required to use the nonresonant cavity for maximizing the absorbed
laser power. Using coherent pumping, it becomes possible to reach
74.1 K in a time of 5 s. Since the optical cooling occurs at the electron
transitions between the SSSs, the limiting minimum temperature of
cooling in the above-developed model is the same as in other anti-
Stokes fluorescence (ASF) models [6].
This investigation shows importance of the nonadiabatic effects
for the problem of the laser cooling of solids doped by RE ions.
The vibronic interaction model used in our calculations specifies
the mechanism of interaction between the ion electrons and ion
vibrations in the crystal matrix. And the density matrix formalism
allows to describe coupling of the ion vibrations with the vibrations
of host material by introducing a mechanism of heat loss. In the
considered case, the loss is attributed to the cooling process. In

addition, the establishment of the Boltzmann distribution at SSSs can
be easily explained by the coupling between the vibronic subsystem
and the phonon reservoir.
As we consider a multilevel system that resonantly interacts with
photonic and vibronic subsystems, the density matrix formalism is
applied. In the framework of this formalism we use the statistical
approximation that implies that the system temperature is constant
throughout the laser cooling cycle. However, the number of quanta
of pseudolocal vibrational modes decreases in that time interval.
Despite the fact that our approach takes into account the temperature
dependence of the quantities that determine cooling performance, a
more rigorous approach to thermodynamics of quantum-mechanical
systems calls for the nonequilibrium statistical operator method
[30].
In the above calculations, no account is taken of the temperature
dependencies of the parameters involved in Eqs. 7.4 and 7.9 and their
solution (the energy separations between levels, the decay rates, and
the vibrational frequencies of the host matrix). However, using the
temperature dependence of these parameters, obtained theoretical
estimations can be only specified but cannot be substantially
changed.
Since VUV radiation is efficiently absorbed in the extremely short
length, the proposed model may be utilized for laser cooling of small
solid-state objects that are used in up-to-date optoelectronics. Also,
this model can be helpful in constructing logic elements of quantum
computers.
References

Nature, 377, pp. 500–503.
T. R. (1997). Laser cooling of a solid by 16 K starting from room
temperature, Phys. Rev. Lett., 78(6), pp. 1030–1033.
3. Epstein, R. I., Brown, J. J., Edwards, B. C., Gibbs, A. (2001). Measurements
of optical refrigeration in ytterbium-doped crystals, J. Appl. Phys.,
90(9), pp. 4815–4819.
References 237
R. (2008). Anti-Stokes luminescence cooling of Tm3+ doped BaY2F8,
Opt. Express, 16(3), pp. 1704–1710.
5. Nemova, G., Kashyap, R. (2010). Laser cooling of Er3+-doped solids,
Opt. Commun., 283, pp. 3736–3739.
6. Seletskiy, D. V., Hehlen, M. P., Epstein, R. I., Sheik-Bahae, M. (2012).
Cryogenic optical refrigeration, Adv. Opt. Photonics, 4, pp. 78–107.
7. Fernandez, J., Garcia-Adeva, A. J., Balda, R. (2012). Anti-Stokes laser-
induced cooling in rare-earth doped low phonon materials, Opt. Mater.,
34, pp. 579–590.
8. Ghasemkhani, M., Albrecht, A. R., Melgaard, S. D., Seletskiy, D. V.,
Cedeberg, J. G., Sheik-Bahae, M. (2015). Intracavity-enhanced optical
refrigeration of Yb:YLF crystal to cryogenic temperatures, Proc. SPIE,
9380, p. 938003.
9. Fleischhauer, M., Imamoglu, A., Marangos, J. P. (2005).
Electromagnetically induced transparency: optics in coherent media,
Rev. Mod. Phys., 77, pp. 633–673.
10. Hemmer, P. R., Turukchin, A. V., Shahriar, M. S., Musser, J. A. (2001).
Raman-exited spin coherences in nitrogen-vacancy color centers in
diamond, Opt. Lett., 26(6), pp. 361–363.
11. Zhao, Y., Wu, C., Ham, B. S., Kim, M. K., Awad, E. (1997). Microwave
induced transparency in Ruby, Phys. Rev. Lett., 79(4), pp. 641–644.
12. Kolesov, R. (2005). Coherent population trapping in a crystalline solid
at room temperature, Phys. Rev., 72, p. 051801(R).
13. Rand, S. C. (2013). Raman laser cooling of solids, J. Lumin., 133, pp.
10–14.
14. Ivanov, A. V., Rozhdestvensky, Y. V. (2015). Laser cooling of doped
crystals by methods of coherent pumping, Proc. SPIE, 9380, pp.
93800S (1-10).
15. Ivanov, A., Rozhdestvensky, Y., Perlin, E. (2015). Coherent pumping for
fast laser cooling of doped crystals, J. Opt. Soc. Am. B, 32(5), pp. B47–
B54.
16. Perlin, Y. E., Kaminskii, A. A. (1985). Nonradiative transitions of the
trivalent lanthanides in insulating laser crystals, Phys. Status Solidi B,
132, pp. 11–40.
17. Lo, D., Makhov, V. N., Khaidukov, N. M., Krupa, J. C., Gesland, J. Y. (2004).
Upconverted VUV luminescence of Nd3+ and Er3+ doped into LiYF4
crystals under XeF-laser excitation, J. Lumin., 106, pp. 15–20.
18. Fischer, G. (1984). Vibronic Coupling (Academic, San Diego).

19. Bersuker, I. B., Polinger, V. Z. (1989). Vibronic Interactions in Molecules
and Crystals (Springer, Berlin).
20. Perlin, Y. E., Wagner, M. (eds.) (1984). The Dynamical Jahn-Teller Effect
in Localized Systems (North-Holland, Amsterdam).
21. Perlin, Y. E., Tsukerblat, B. S. (1974). Electron-Vibrational Interaction in
Optical Spectra of Impurity Paramagnetic Ions (Shtintsa, Kishinev) (in
Russian).
22. Saiko, A. P., Kuz’min, V. S. (1997). Lasing and Dicke superradiation by
vibronic transitions in impurity crystals, Laser Phys., 7(2), pp. 384–
386.
23. Vaikjärv, T., Hizhnyakov, V. (2014). Time-dependent pseudo Jahn-Teller
effect: phonon-mediated long-time nonadiabatic relaxation, J. Chem.
Phys., 140, p. 064105.
24. Klauder, J. R., Sudarshan, E. C. G. (1968). Fundamentals of Quantum
Optics (Benjamin, New York); Glauber, R. J. (2007). Quantum Theory
of Optical Coherence. Selected Papers and Lectures (Wiley-VCH,
Weinheim).
25. Vitanov, N. V., Halfmann, T., Shore, B. W., Bergmann, K. (2001). Laser-
induced population transfer by adiabatic passage techniques, Annu.
Rev. Phys. Chem., 52, pp. 763–809.
26. Leniec, G., Kaczmarek, S. M., Boulon, G. (2005). EPR and optical
properties of CaF2:Yb single crystals, Proc. SPIE, 5958, p. 595825.
27. Szczurek, T., Schlesinger, M., Jezowska-Trzebiatowska, B.,
Legendziewicz, J., Strek, W. (eds.) (1985). Rare Earth Spectroscopy
(World Scientific, Singapore).
28. Misra, P., Dubinskii, M. A. (eds.) (2002). Ultraviolet Spectroscopy and
UV Lasers (Marcel Dekker, New York).
29. Edwards, B. C., Anderson, J. E., Epstein, R. I., Mills, G. L., Mord, A. J.
(1999). Demonstration of a solid-state optical cooler: an approach to
cryogenic refrigeration, J. Appl. Phys., 86(11), pp. 6489–6493.
30. Andrianov, S. N., Samartsev, V. V. (1998). Laser cooling of impurity
crystals, Laser Phys., 8(1), pp. 14–18.
Chapter 8
Bulk Cooling Efficiency Measurements

of Yb-Doped Fluoride Single Crystals and
Energy Transfer–Assisted Anti-Stokes
Cooling in Co-Doped Fluorides

Dipartimento di Fisica-NEST Instituto di Nanoscienze CNR,
Universita`di Pisa, L. B. Pontecorvo 3, I-56127, Pisa, Italy
volpi@df.unipi.it
8.1 Introduction
Optical cooling of solids, based on anti-Stokes fluorescence, has
tremendously advanced in the last years. Since its first experimental
demonstration in 1995 [5], the temperature drops achieved have
dramatically increased from a few degrees up to more than 150.
Operations in the cryogenic regime have been reached in ytterbium
(Yb)-doped fluoride single crystals [13, 14, 21]. To date, the lowest
temperature achieved is 93 K [12] and novel schemes based on
energy transfer processes prospect significant enhancements

www.panstanford.com
240 Bulk Cooling Efficiency Measurements of Yb-Doped Fluoride Single Crystals
of intrinsic efficiencies, pushing down the minimum achievable

temperatures.
Truly vibration-free, cryogenic laser cooling is a much sought
after solution for several space-based technologies. The use of
mechanical coolers indeed pushes to complex system solutions
in order to minimize microphonic vibrations, which limit the
detectors’ performances. Accessing temperatures well below the
thermoelectric coolers’ (TECs) cutoff (170 K), the implementation of
cryogenic optical coolers would provide an ideal solution for cooling
down detectors in the temperature range below 170 K. Moreover,
optical coolers would offer an efficient alternative to TECs in the
higher temperature range, below 200 K, where TECs exhibit limited
wall-plug efficiency. Currently, passive radiators, relying on thermal
radiative emission into deep space, provide the main vibration-
free solution for cooling down detectors below 170 K, down to 50
K. The use of these devices, however, imposes several orientation
constraints and shielding requirements to build up the refrigeration
structure. Furthermore large emitting surfaces are necessary to
achieve low temperatures and relatively high cooling powers. The
radiated power, and hence the available cooling power, is directly
proportional to the panel surface as well as to the fourth power of
the temperature to reach. Consequently, due to size limitations, the
performances rapidly decrease as the temperature to achieve drops,
especially below 100 K. The optical cooling technology offers, instead,
advantages of extreme compactness and cooling efficiencies almost
independent of dimensions, enabling significant size and weight
reduction. In addition to that, optical cooling allows for simpler
designs, without orientation and shielding constraints, and possesses
a much higher potential for intrinsic efficiency enhancement
through the development of novel materials. Additional advantages
for space applications, offered by the optical cooling technology, are
a long lifetime (only limited by the lifetime of the pumping laser),
low electromagnetic interference, low sensitivity to magnetic fields,
and extremely high reliability.
All these features intensely motivated the research toward device
implementation as well as investigation in material development to
uplift the intrinsic efficiency of cooling crystals, which are currently
of the order of a few percent from room temperature. Various
techniques have been exploited and many are under investigation.
Laser cavities were introduced to enhance the absorption of
Optical Cooling Model for Rare Earth–Doped Materials 241
incident laser photons and so the available cooling power at low

temperatures [12–14, 21]. Suitable configurations were exploited
to reduce parasitic heat loads. Novel alternative schemes based on
photon localization effects in nanocrystalline powders were recently
suggested to promote the input absorption [19]. The superirradiance
regime was proposed to increase the cooling process rate [15].
While several methods have been proposed and can be employed
to improve the cooling performances, energy transfer–assisted
anti-Stokes processes in co-doped systems have been recently
demonstrated to provide a route to enhance the intrinsic cooling
efficiency of active materials. The first experimental evidence was
achieved with Yb–Tm co-doped systems. Efficiency enhancements,
over the single Yb doping performances, have been demonstrated
via co-doped YLF single crystals [24]. Preliminary results achieved
in our laboratory are presented in the following section.
This chapter reports on cooling efficiency measurements
performed on various Yb-doped fluoride single crystals grown in our
laboratory. The investigation of the cooling process for increasing Yb
concentrations in LiYF4 (YLF) single crystals resulted in improved
cooling efficiency via significant decrease of detrimental parasitic
absorption. Energy transfer processes between Yb ions and low
concentrations of rare earth ion impurities were investigated to
analyze the effect of detrimental impurities that participate in
multiphonon relaxation of excited ions. Studies on Yb–Tm-controlled
co-doping led to the first observations of energy transfer–enhanced
Yb anti-Stokes efficiency [24]. Preliminary results achieved with Yb–
Tm co-doped samples are presented.
The following section introduces the anti-Stokes efficiency
model for rare earth–doped materials. Experimental results and a
description of experimental apparatus and instruments are reported
in Sections 8.3–8.5.
8.2 Optical Cooling Model for Rare

Earth–Doped Materials
The idea of using rare earth ions to achieve anti-Stokes cooling in the
solid phase was firstly proposed by Kastler in 1950 [9]. The interest
for these ions basically relies on their electronic configuration. In
the triply ionized state, indeed, the 4f valence electrons are heavily
shielded from the surrounding crystal field by the filled 5s and

5p shells, which have a larger spatial extension (Fig. 8.1). When
inserted as impurities in a solid, trivalent rare earth ions hence
weakly interact with the crystal field of the surrounding. As a
result, these ions exhibit atomic-like spectra even when embedded
in hosting matrices and the 4f–4f optical transition can be excited
in transparent solids. Such transitions amongst 4f states, parity
forbidden in the free ions, are weakly allowed in hosting crystals
because of induced dipoles due to the interaction of ions with the
crystal field, mixing 4f states with different parities.
1.4
4f 4f
1.3
5s
1.2 5p
6s
1.1
1.0 5s
.9 5p
.8
P2(4) .7
.6
.5
.4 6s
.3
.2
.1
0.2 .6 1.0 1.4 1.8 2.2 2.6 3.0 3.4 3.8 4.2 4.6 5.0 5.4 5.8 6.2 6.6 7.0 7.4
p(au)
Figure 8.1 Radial charge density of the 4f, 5s, 5p, and 6s shells of the Gd3+
free ion. Reprinted with permission from Freeman, A. J. and
Watson, R. E., Phys. Rev. (1893–1969), 127, 1962. Copyright
1962, by the American Physical Society.
The basic advantage of using rare earth ions for optical

refrigeration, however, is related to the extremely high radiative
efficiency achievable with 4f–4f transitions. The efficient screening
of valence electrons strongly reduces the electron–phonon coupling
and suppresses nonradiative decays due to multiphonon emissions
from excited 4f states.
Additionally, the screening of the 4f orbital minimizes the Stokes
shift of the emission associated with the 4f–4f transitions, allowing
states with relatively a large absorption cross section to be utilized
for the anti-Stokes cooling process. Such a process is basically
enabled by the interaction of active ions with the crystal field, which
causes, at the first order of the perturbation theory, a Stark splitting

of the free ion energy levels, with a magnitude of the order of 102
cm–1. Such splitting leads to energy widths for the 4f manifolds of
the order of a few hundreds of cm–1, sufficient for heat removal
via anti-Stokes emission. By exciting the lower-energy inter-Stark
transition, annihilation of lattice phonons occurs to reequilibrate
the Boltzmann distribution of the Stark populations. Fluorescence at
higher average energies than pumping photons follows, carrying out
thermal energy from the system. Thermalization of the electronic
excitations with the phonon field (10–12 s) occurs at a much faster
timescale than spontaneous emission (10–3 s). Bulk cooling of the
solid, however, occurs only if the decay is predominantly radiative.
Loss channels such as nonradiative decay rates or energy transfer
from excited ions thus need be minimized, as well as trapping of
anti-Stokes photons.
In the ideal anti-Stokes process each emitted photon removes
an average energy equal to the difference between the emitted and
the absorbed photons. The cooling efficiency can be defined as the
ratio of the energy absorbed from the lattice to the input energy. The
efficiency of the ideal anti-Stokes process hc id can thus be formalized
as
hn f - hn l
hc id = = -1 (8.1)
hn lf
where vf(λf) is the mean emission frequency (wavelength) of
the electronic transition and v(λ) the frequency (wavelength) of
pumping photons. In real cases, however, in addition to the anti-
Stokes process, heat-generating mechanisms, related to multiphonon
quenching and impurity-mediated absorption, can occur after laser
excitation. Furthermore, fluorescence photons can be trapped due to
reabsorption and internal reflection phenomena.
The currently used model for anti-Stokes efficiency in rare earth–
doped solids, addressing competing radiative and nonradiative
optical processes and loss channels, was developed by Epstein
et al. [22]. In this model the probability of a pumping photon to
be converted into an anti-Stokes photon escaping the material is
quantified by two parameters, the external quantum efficiency
(EQE) of the electronic transition and the background absorption
coefficient of the material. The EQE parameter defines the radiative
efficiency of the electronic transition, addressing the probability of

photon trapping. The background absorption coefficient defines the
actual absorption efficiency, accounting for loss mechanisms that
lower the number of absorbed photons that may initiate the anti-
Stokes process. In this model, the cooling efficiency, dependent on
temperature T and pumping wavelength λ, is formalized as [20, 22]
l
hc ( l ,T ) = hext habs ( l ,T ) -1 (8.2)
lf (T )
where ηext represents the EQE of the electronic transition and ηabs(λ,
T) the absorption efficiency, defined as
a r ( l ,T )
habs ( l ,T ) = (8.3)
a r ( l ,T ) + a b
where ar(λ, T) is the resonant absorption of the electronic transition
and αb the background absorption parameter of the material. Net
cooling requires a positive value of ηc. Such condition imposes
stringent requirements on the value ηextηabs, that is, on the material
efficiency. Considering, at temperature T, a detuning of magnitude
kBT between emitted and absorbed photons, net cooling (ηc > 0)
requires
kBT
hext habs > 1 - (8.4)
hn f
Considering a transition gap of 1 eV, such as for the Yb transition,
ηextηabs > 98% is required to achieve net cooling starting from room
temperature. If the temperature is reduced to 100 K, the efficiency
requirement increases to ηextηabs > 99% for cooling to occur. Both
an EQE close to unity and minimized background absorption are
thus fundamental conditions to achieve efficient cooling. Stringent
requirements on the crystal quality follow.
It is well known that the EQE parameter is basically related
to the probability of nonradiative decay and to the efficiency of
photon extraction. In the formula next = ηeW/ηeWr + Wnr, ηe is the
fluorescence extraction efficiency and Wr and Wnr the radiative and
nonradiative decay rates, respectively, of the electronic transition.
Fluorescence photons basically get trapped inside the material due
to reabsorption and total internal reflection, while nonradiative
decays are basically related to multiphonon relaxations and phonon-
assisted energy transfers.
The choices of electronic transition and hosting material are

fundamental to minimize the efficiency of nonradiative quenching
and photon-trapping phenomena. Optical transitions of trivalent
rare earth ions, embedded in fluoride hosts, are preferable as
they allow for high EQE values due to the low electron–phonon
coupling and low cutoff of optical phonon energy of fluoride
crystals. Furthermore the low refractive index of fluoride crystals
allows for high-fluorescence escape efficiencies, reducing photon
trapping due to total internal reflections. Among trivalent rare
earths, however, Yb3+ ions possess superior properties that enable,
when embedded in fluoride hosts, extremely reduced nonradiative
decay rates and hence exceptionally high EQE values. Firstly, the
absence of a higher excited state, above the first excited manifold
2F
5/2, hinders nonresonant Yb–Yb energy transfer processes, which
are likely for rare earth ions to cause efficient quenching of excited
states via multiphonon release. The absence of higher excited states
also enables the use of high concentrations of Yb ions, to enhance
absorption efficiency, without foster detrimental energy transfer
processes. Secondly the large energy gap of the Yb transition, around
1 µm, inhibits multiphonon relaxation processes, reaching negligible
rates in fluoride hosts. Furthermore, high-power pump sources (>10
W CW) emitting around 1 µm are readily available. All these features
contributed to the extremely high cooling performances achieved in
Yb-doped fluoride materials.
Currently, the most critical parameter limiting the efficiency of
such materials is the background absorption coefficient, αb. This
parameter reckons with all the parasitic processes competing
with the resonant absorption of the electronic transition. Although
its nature and origin are still under investigation, background
absorption is basically related to unwanted impurities embedded
within materials during the synthesis. Optically active contaminants
introduce additional energy levels, which are likely to cause
absorption of pumping photons, phonon-assisted energy transfer,
and multiphonon quenching of electronic excitations. Spurious
laser scattering due to crystal defects also contributes to parasitic
absorption, violently heating up the lattice. All these processes
significantly degrade the absorption efficiency of the anti-Stokes
process, reducing the number of pumping photons that can initiate

the anti-Stokes process. Nonradiative impurity-mediated processes
also degrade the EQE of the electronic transition. The cooling
mechanism can even be compromised if the efficiency of the heat-
generating processes is not balanced by the efficiency of the anti-
Stokes cooling process [11]. Several studies identify the severe
impact of transition metals on the Yb anti-Stokes efficiency [7, 14].
Divalent ions of 3d metals, such as Cu, Fe, Co, and Ni, have strong
absorption bands around 1 µm and thus efficiently absorb pump
power, lowering the number of pump photons that can initiate the
anti-Stokes process. A large background absorption decrease was
observed in YLF crystals with the reduction of Fe2+ ions in cooling
samples, resulting in net efficiency enhancement. Rare earth
impurities were also observed to influence the efficiency of the Yb
anti-Stokes process, degrading the EQE via nonradiative relaxation
and heat-generating energy transfer processes [3]. Any other kind
of impurities that participate in nonradiative quenching of excited
ions degrades the anti-Stokes efficiency, leading to poor cooling
performances. Calculations indicate that further purification of
starting materials, leading to a significant background absorption
decrease, is the way to improve the current state-of-the-art cooling
efficiencies, pushing down minimum achievable temperatures,
while increasing the cooling powers available at low temperatures.
Due to the chemical inertness of binary fluorides, however, the
control of impurities in binary fluorine materials is quite complicate
and the purification below the part-per-million level is still a quite
challenging process. One approach currently under investigation is
provided in Ref. [6].
The background absorption coefficient and EQE parameters are
normally used to quantify the cooling performances of investigated
materials. Such parameters can be estimated from the fit of
experimental data of cooling efficiency with the model curve in Eq.
8.2, where data of the mean emission wavelength and resonant
absorption coefficient need to be inserted for each material. Such
spectroscopic parameters need to be previously determined from
optical investigations of the electronic transition, through absorption
and emission measurements.
Experimental Setup 247
8.3 Experimental Setup

Before performing cooling efficiency measurements, the crystalline
active materials were spectroscopically characterized by absorption
and static and dynamic fluorescence measurements. In this
section the experimental apparatus and instruments employed for
spectroscopic and cooling measurements are described.
8.3.1 Spectroscopic Setup

Absorption measurements were performed by means of a dual-
beam design spectrometer Varian Cary 500, operating between 180
nm and 3200 nm. The absorbance range of the machine is 0–10 with
resolution 0.001. The nominal wavelength resolution is 0.01 nm in
the UV-visible range and 0.04 nm in the IR. Polarized absorbance
spectra were acquired by placing two identical Glam–Thompson
polarizers on the reference and sample paths in order to linearly
polarize the beam incident onto the sample.
To measure the fluorescence spectra, the sample was excited
by means of a continuous-wave (cw) laser at a fixed wavelength
and the intensity of the emission was registered as a function of
the wavelength. In the case of a large overlap between absorption
and emission bands, such as for the Yb transition, reabsorption
phenomena can cause significant distortions of the intrinsic
fluorescence spectra. At shorter wavelengths, where strong
absorption is also present, the emissions peaks tend to be decreased
as the emitted radiation is partially reabsorbed before it exits the
material. Depleted fluorescence is hence observed in this region.
To minimize distortions of the observed fluorescence due
to reabsorption-trapping artifacts, the sample was mounted on
a pinhole support, which enables one to minimize the overlap
between the volume of the sample that is excited and the volume of
the sample from which fluorescence is observed. At the same time,
the emission signal can be acquired perpendicularly to the excitation
beam in order to minimize spurious laser scattering. Polarized
pumping can be accurately performed as well as observation of
selective fluorescence from each optical axis. The fluorescence from
the sample, after laser excitation, was modulated by using a rotating
chopper, with 77 Hz frequency, and focused on the entrance slit of a
monochromator equipped with interchangeable gratings (300, 600,

1200 grooves/mm) and input and output slits adjustable between
0.01 and 2 mm. The fluorescence signal was then acquired by using
a suitable detector placed at the exit of the monochromator. For
IR fluorescence the monochromator was equipped with the 600
grooves/mm grating (with dispersion of 5.4 nm/mm) and a 0.1
mm slit aperture; the signal was acquired by means of an indium
antimonide (InSb) photodiode (EG&G Judson) cooled to 77 K with
liquid nitrogen. The working range of the detector extends from 1 to
5.5 µm and the sensitivity is 2.5 mA/W at 5 µm. For the visible–near-
IR (NIR) region the monochromator was equipped with the 1200
grooves/mm grating (with dispersion of 2.77 nm/mm) and a 0.2 mm
slit aperture; the signal was detected by means of a photomultiplier
tube (Hamamatsu R943-02). Its spectral response ranges from
160 nm to 930 nm, with a sensitivity of 65 mA/W at 82 nm. The
output signal from the detector was processed by a lock-in amplifier
that used the chopper frequency as reference and stored on a PC.
Polarized spectra were measured by placing a Glam–Thompson
polarizer in front of the entrance slit of the monochromator. Raw
data acquired were calibrated for the optical response of the
apparatus, collecting the fluorescence spectra of an incandescent
lamp with emission spectra equivalent to a black body at 3000 K.
The optical response curve of the apparatus was derived combining
acquired fluorescence spectra from the lamp and theoretical black-
body emission spectra.
For dynamic fluorescence measurements, the emission temporal
decay was investigated after laser excitation of the sample with
a pulsed source. For the Yb transition, the pinhole method [10]
was used to minimize lengthening of the intrinsic lifetime due to
reabsorption phenomena. To implement such a method, the sample
was mounted behind a pinhole mask and the temporal decay of the
fluorescence was measured as a function of the aperture radius.
The intrinsic radiative lifetime was then extrapolated as the value
corresponding to a zero hole radius [10]. This method was not
used for static fluorescence measurements, because acquisition of
polarized spectra was necessary as well as polarized pumping along
a single optical axis.
The excitation source for dynamic measurements was a tunable
titanium-Al2O3 (Ti:Sa) pulsed laser with an emitting wavelength
Experimental Setup 249
in the range between 750 nm and 870 nm and a maximum pulse

energy of 9 mJ. The pulse duration was 30 ns and the repetition rate
10 Hz.
For measurements of the Yb transition lifetime, the Ti:Sa
beam was tuned to a maximum in the absorption (~960 nm). The
fluorescence from the sample was focused on the entrance slit of the
monochromator, equipped with the 600 grooves/mm grating and
locked at a fixed wavelength, corresponding to a maximum emission
peak. The fluorescence signal was then acquired by means of an
InSb photodiode. The signal from the detector was amplified, by
fast amplifiers, and then processed by using a digital oscilloscope,
triggered at the pulse repetition frequency of the laser excitation
source.
8.3.2 Cooling Setup

Cooling efficiency measurements were performed in a single-pass
configuration by monitoring the heating/cooling temperature
change of the sample after laser excitation at varying wavelengths
within the Yb absorption spectrum. A layout of the cooling setup is
shown in Fig. 8.2. The sample was mounted inside a steel vacuum
chamber (10 × 10 × 10 cm3), evacuated at a pressure of 10–5 Pa,
in order to minimize convective heat load on the sample. Inside
the vacuum chamber, the sample was mounted on a stainless steel
support suspended on two optical fibers of 120 µm diameter. Such
a scheme was designed in order to minimize parasitic conductive
heat loads, ensuring thermal isolation of the sample from the
surrounding. Four windows on the chamber walls enabled excitation
and acquisition of the sample temperature simultaneously.
Four diode laser sources, with central wavelengths between
940 and 1060 nm, were used to excite the sample. All the junctions
were mounted to excite the sample along the c axis, which allows the
highest absorption coefficient. Antireflection-coated BK7 windows
were used to pass the laser beams trough the vacuum chamber. To
measure the temperature of the sample, contactless thermometry
techniques were used. Two methods were employed, a thermal
camera with a microbolometer sensor and a spectroscopic setup
for acquisition of fluorescence spectra from the sample, which allow
one to implement the differential luminescence technique (DLT).
THERMAL
SAMPLE VACUUM CHAMBER
CHAMBER
BaF2 window Thermocouple
SAMPLE
Mirror
Optical fibers
f = 10 cm f = 10 cm LD(1030 nm) (d = 120 mm)
glass window
BK7 windows
AR 1030 nm
LD (980 nm)
LD (1060 nm)
LD (940 nm)
SPECTROMETER
PC
optical fiber
Figure 8.2 Layout of cooling setup for cooling efficiency measurements.

On the right is shown a sketch of the sample mounting inside
the vacuum chamber.
A long-wavelength infrared (LWIR) thermal camera (Raython

2500 AS) with a photoconductive sensor, made of an uncooled
amorphous silicon microbolometer array, was used for the smaller
temperature changes, below 5 K. The output of the camera is a gray-
scale image file with 8-bit pixels, where dark pixels correspond to
hotter areas, while colder areas show up as lighter pixel regions. The
nominal resolution of is 0.2 K. A BaF2 window, which is transparent
up to 10 µm, was used to allow the thermal sensor to acquire the
fluorescence from the sample. For each pumping wavelength, a
thermal video was registered while the sample was excited with the
laser beam. The video started with the absence of laser excitation
and extended until the thermal equilibrium was reached. From the
images acquired, the temperature change of the sample with respect
to the surrounding was derived. Thermal images were acquired
at a rate of 0.1 Hz. The stability of the chamber temperature was
monitored by means of a suitable thermocouple. The thermal camera
was previously calibrated in order to determine the proportionality
relationship between the pixel values and the temperature of the
sample. A linear working range of ±5 K was measured.
For larger temperature changes the DLT was used [21]. Such a
technique is based on changes in the shape of the emission spectra as
a function of temperature. As the temperature lowers, the decrease
of phonon excitations reflects in changes of the electronic excitations
distribution of the Stark sublevels. The vibronic broadening of the
electronic emission spectra decreases as a function of temperature;
consequently the emission lines narrow and become more intense.
Experimental Results 251
The DLT uses the difference between integrated areas of emission to

measure the temperature changes.
Emission spectra from the sample, excited with laser beams
inside the vacuum chamber, were acquired from a glass window. As
for the thermal camera, the acquisition of the spectra started with the
absence of laser excitation and extended until thermal equilibrium
was reached. A fiber-coupled spectrometer, operating between 200
and 1100 nm with 1 nm resolution, was used. A previous calibration
was performed to connect the differential spectrum signal to the
temperature value. For the calibration measurements, the cooling
setup geometry was closely resembled in order to cause similar
distortion effects due to reabsorption. The temperature of the sample
was varied by using a Peltier cell and the fluorescence spectra were
acquired as a function of the temperature. The estimated resolution
for the DLT was of the order of 1 K.
8.4 Experimental Results

This section reports on results of cooling efficiency measurements
performed on Yb-doped YLF single crystals grown in our facility.
The EQE and background absorption parameters were evaluated
from experimental data of cooling efficiency and outcomes of
spectroscopic analysis.
The effect of increasing concentrations of Yb ions on the anti-
Stokes efficiency was investigated. Larger concentrations of Yb ions
resulted in efficiency enhancements due to diminishing background
absorption. Different relative concentrations of rare earth impurities
were observed to result in depleted EQE values in the case of higher
amounts of specific rare earths ions in certain samples. Energy
transfer processes between Yb and low concentrations of impurities
were investigated. Significant efficiency enhancements, over the
single Yb doping performances, were achieved in Yb–Tm co-doped
YLF single crystals, providing the first experimental evidence
of energy transfer–assisted anti-Stokes cooling. Preliminary
results prospect appreciable improvements for low-temperatures
operations, besides huge potentialities for intrinsic efficiency
enhancements of active materials.
Prior to cooling experiments, spectroscopic measurements
were performed to evaluate the resonant absorption coefficient
and the mean emission wavelength of the Yb transition. Bulk

cooling experiments were performed in a single-pass configuration
measuring the heating/cooling temperature change as a function of
the pumping wavelength.
8.4.1 Investigation of Anti-Stokes Cooling in YLF Single

Crystals with Varying Yb Doping Levels
The anti-Stokes efficiency was systematically studied in YLF
crystalline hosts with varying Yb doping levels. Single crystals doped
Yb at 5 at%, 7.5 at%, and 10 at% were investigated.
The increase of active ion concentration basically results in
enhancement of the absorption coefficient of the material, which
is particularly sought after in the excitation region for the anti-
Stokes process, at λ > λf. This region indeed corresponds to the
long-wavelength tail of the absorption spectrum, where resonant
absorption is low. Enhancements of Yb concentration, increasing
the resonant absorption, enable improvements of the absorption
a r ( l ,T )
efficiency for the anti-Stokes process: habs ( l ,T ) = .
a r ( l ,T ) + a b
Intensification of reabsorption-related trapping phenomena,
however, needs to be considered when the active ion concentration
is increased.
Prior to cooling experiments, absorption and static and dynamic
fluorescence measurements were performed for each doping level
in order to calculate the resonant absorption coefficient and the
mean emission wavelength and, furthermore, in order to check for
anomalies.
Figure 8.3a shows the polarized absorption spectra acquired for
the YLF:5%Yb sample at room temperature. The multipeak structure
of the spectra is related to the inter-Stark transitions between Stark
sublevels of the two manifolds. The highest absorption, over the
whole spectrum, is achieved through selected excitation along the c
axis. The strongest peak in the YLF:5%Yb is 8.36 cm–1 at 959.6 nm.
Similar behaviors were observed for the higher Yb doping levels.
The spectra showed a similar structure but increasing magnitude.
In Fig. 8.3b are reported the absorption spectra of the 2F5/2 excited
manifold acquired, in the same experimental conditions, for the three
doping levels (5%, 7.5%, and 10%) in π-polarization. As can be seen
from the graph, the magnitude of the resonant absorption increases
linearly with the Yb concentration, leading to improvement of about
a factor of 2 between 5% and 10% Yb doping in the excitation region

for anti-Stokes.
E || c
E^c
(a)
YLF:10% Yb
YLF:7.5% Yb
YLF:5% Yb
(b)
Figure 8.3 (a) Polarized absorption coefficient of YLF:5%Yb acquired.
(b) Absorption coefficient of YLF:5%Yb, YLF:7.5%Yb, and
YLF:10%Yb for π-polarization.
Following excitation of the 2F5/2 manifold, the spontaneous

emission yields a broadband fluorescence spectrum I(λ) with a
mean wavelength of
lf =
Ú I(l )ldl (8.5)
Ú I(l )dl
An excitation tuned to wavelengths longer than λf is required to
achieve anti-Stokes emission.
Polarized fluorescence spectra were acquired for the 5%, 7.5%,
and 10% Yb-doped YLF samples after laser excitation of the Yb
2F
5/2 manifold at 940 nm. Fluorescence spectra registered for the
YLF:5%Yb and YLF:10% samples are shown in Fig. 8.4.
Although the structure of emission spectra does not change
as the Yb concentration increases, the increase of reabsorption-
related effects for larger concentrations of active ions influences the
intensity distribution of the emission.
Due to the large overlap between absorption and fluorescence
bands, photons that are spontaneously emitted by excited ions
tend to be efficiently reabsorbed by other ions in the ground
state. Increasing the concentration of Yb ions, radiation-trapping
phenomena, due to reabsorption, tend to be more efficient, leading
to more depleted emissions at higher energies, as can be observed
by comparing spectra in Fig. 8.4.
Fluorescence spectra were used both to calculate the mean
emission wavelength parameter and the absorption coefficient via
the reciprocity method [17]. This technique is used to improve the
accuracy of the absorption coefficient measurement in the long-
wavelength tail of the spectrum, that is, the excitation region for
anti-Stokes emission. Due to the low value of the absorption in this
region, direct measurements normally exhibit large noise. Using
reciprocity theory, instead the resonant absorption is derived from
fluorescence spectra. The emission spectra exhibit a considerably
higher signal-to-noise ratio in this wavelength region and hence
improved accuracy for the absorption coefficient evaluation. Figure
8.5 reports, for comparison, the absorption coefficient of YLF:5%Yb
for the π-polarization calculated from direct measurements and
by using reciprocity. As can be seen from the graph, the reciprocity
theory well reproduces the behavior of the absorption coefficient
with significantly higher accuracy.
E || c
E^c
(a)
E || c
E^c
(b)
Figure 8.4 Polarized fluorescence spectra of YLF:5%Yb (a) and
YLF:10%Yb (b), acquired, between 920 and 1100 nm with a
resolution 0.5 nm, after laser excitation of Yb ions at 940 nm
along the c axis.
Figure 8.5 Absorption coefficient of YLF:5%Yb calculated from direct

measurements (blue data points) and by using reciprocity
theory, starting from fluorescence spectra (red data points),
for π-polarization.
Dynamic fluorescence measurements, performed by means of

the pinhole method (see Section 8.3.1), yielded, for all the three
doping levels, nearly the same value for the Yb radiative lifetime, τ =
2.1 ± 0.2 ms, in agreement with literature data [2].
Static and dynamic fluorescence measurements were used to
calculate the β-τ emission cross sections required to implement
the reciprocity method. In Fig. 8.6 are shown the β-τ emission cross
sections calculated for the YLF:5%Yb and the YLF:10%Yb samples.
From the fluorescence spectra acquired, the mean emission
wavelength of the Yb transition was calculated by taking the first
moment of the fluorescence function (Eq. 8.5). For increased Yb
doping levels, a slight red shift of the λf value was observed. Such
an effect can be ascribed to the increase of reabsorption trapping
phenomena, which strengthen with the increase of the active ion
concentration. The spectroscopic parameters of the mean emission
E || c
E^c
(a)
E || c
E^c
(b)
Figure 8.6 Polarized β-τ emission cross sections of YLF:5%Yb (a) and
YLF:10%Yb (b).
wavelength and the resonant absorption coefficient obtained from

spectroscopic measurements were used to build the model curve of
cooling efficiency in Eq. 8.2. Experimental data of cooling efficiency

were acquired by measuring the heating/cooling temperature
changes of the sample as a function of pump photon energy. The
apparatus used for these measurements is detailed in Section
8.3.2. Basically, the sample was placed inside a vacuum chamber,
suspended on two optical fibers of 120 µm diameter, in order to
minimize conductive heat loads. The chamber was evacuated at low
pressure (10–5 Pa) in order to minimize convective heat transfer.
Four different laser sources, with a central wavelength between 940
nm and 1060 nm, were used to excite the sample along the c axis. For
each pumping wavelength, the temperature change was measured
while the sample was excited with the laser beam, starting from
the absence of laser irradiance until thermal equilibrium, under
laser pumping, was reached. Figure 8.7 shows the typical temporal
evolution of the temperature, under laser excitation in the anti-
Stokes region.
Figure 8.7 Temporal evolution of temperature, and relative exponential

theoretical fit, for the YLF:5%Yb sample under laser excitation
at 1024 nm with 1.5 W of pump power.
Data analysis confirmed the exponential change of the sample

temperature, as expected from thermodynamics.
Assuming black-body radiation as the dominant heat load,

the cooling efficiency results were proportional to the ratio of the
temperature change to the absorbed power. Experimental data
of cooling efficiency were measured for all the available pumping
wavelengths. The experimental data of cooling efficiency were fitted
with the model curve (Eq. 8.2). From the fit procedure, the EQE and
background absorption parameters were evaluated for each sample.
Results obtained for the three different Yb doping levels are reported
in Table 8.1 [24].
Table 8.1 EQE and background absorption values estimated for YLF
single crystals with varying Yb doping levels
EQE αb (104 cm–1)

YLF:5%Yb 0.992 ± 0.001 4.0 ± 0.4
YLF:7.5%Yb 0.983 ± 0.001 2.7 ± 0.4
YLF:10%Yb 0.982 ± 0.001 2.1 ± 0.5
Large efficiency enhancements were observed for increasing

Yb concentrations, due to a significant decrease of background
absorption. Figure 8.8 reports experimental data points and a
relative model curve for the 5%, 7.5%, and 10% Yb-doped YLF
samples.
As can be seen from the graph, the peak efficiency increases
with the Yb concentration due to decreased background absorption
losses. A slight depletion of the EQE, however, is observed for larger
Yb concentrations. Such an effect is ascribable to the enhancement
of reabsorption-related phenomena, which red-shift the mean
emission wavelength of the electronic transition.
Data collected result in good agreement with those reported
in Ref. [4], providing good validation of our experimental setup,
besides indication of the high optical quality of our active materials.
Further increases of Yb concentration in YLF crystals are currently
under investigation. The balance between detrimental reabsorption
effects and virtuous resonant absorption enhancements needs to be
previously studied.
Although the total concentration of contaminants inside cooling
crystals was tested at the 5N level, different relative concentrations
of rare earth impurities at the part-per-million level were observed
to cause appreciable variations of cooling efficiency [3]. Two YLF

single crystals both doped with 5% Yb but containing different
relative concentrations of impurities, due to the use of different
raw material batches, exhibited dissimilar cooling performances.
Although both samples possessed total amounts of impurities
below 10 ppm, a depleted EQE of a few tenth % was measured in
one sample, containing higher amounts of Er and Ho impurities. The
investigation of energy transfer processes in these samples showed
Er and Ho impurities to induce significant heat dissipation, via
multiphonon emission in internal processes and direct transfer of
excitation from Yb ions [3]. However, while some impurities, such
as Er and Ho ions, appear to induce detrimental heat generation in
energy transfer from Yb ions, other rare earth impurities can result
in virtuous processes, energetically favorable for the Yb anti-Stokes
process. Motivated by previous studies on the effect of rare earth
impurities on the Yb anti-Stokes efficiency, the Yb–Tm co-doping was
investigated. Efficiency enhancements via energy transfer–assisted
anti-Stokes cooling were achieved. Results are reported in the
following section.
Yb10%
Yb5%
5%
7.
Yb
Figure 8.8 Experimental data points and fit model curve of cooling
efficiency for Yb-doped YLF samples of variable doping levels.
8.4.2 Yb–Tm Energy Transfer–Enhanced Anti-Stokes

Efficiency in Yb-Doped YLF Single Crystals
A novel scheme for anti-Stokes efficiency enhancement, based on
energy transfer processes, has been investigated in co-doped YLF
crystals [24]. Energy transfer induced by impurities typically results
in loss channels, reducing the anti-Stokes efficiency via multiphonon
quenching of excited ions, hence providing heat release in the lattice.
In our work, we identified a virtuous energy transfer mechanism,
based on Yb–Tm co-doping, which resulted in efficiency enhancement
of the Yb anti-Stokes process via cooperative sensitization of Yb
pairs, followed by radiative emissions at average higher energies
than pumping photons, without phonon release.
A theoretical scheme for anti-Stokes efficiency enhancement via
energy transfer processes was firstly suggested by Dong et al. [4]
in 2013 for a Tm–Ho co-doped system. In such scheme, the energy
transfer induced by interaction between Ho ions excited to the 5I7
state and Tm ions in the ground state is employed to enhance the
efficiency of the Ho anti-Stokes process around 2 µm. Via Ho de-
excitation 5I7 Æ 5I8, few Tm ions get sensitized to the 3F4 state by
absorbing the energy transferred from Ho excitations. Radiative
emission from the 3F4 state, which occurs at higher energy than
Ho anti-Stokes emission, that is, higher than pumping photons,
provides an extra release of thermal energy from the material. Such
a mechanism, however, strongly suffers from detrimental energy
transfer to upper energy levels. Moreover relatively high-power
pumping sources around 2 µm are not readily available. As far as we
know, these kinds of co-doped systems have been only theoretically
investigated.
The Yb–Tm system instead exhibits several advantages. First
of all, Yb-doped fluorides currently enable the best laser cooling
performances, and efficiency enhancements via Yb–Tm co-doping
appear promising to achieve significant improvements of current
efficiencies, pushing down the minimum achievable temperature,
while uplifting the available cooling powers at low temperatures.
Furthermore, due to the respective energy level structure of the two
ions, for specific relative concentrations, Yb–Tm endothermic energy
transfer can occur via cooperative sensitization of Yb ions, providing
an extra release of thermal energy from the lattice.
The effect of Yb–Tm energy transfer on the Yb cooling efficiency

was investigated for low Tm doping levels. Net enhancement of
anti-Stokes efficiency was achieved via Yb–Tm energy transfer in co-
doped YLF single crystals, providing the first experimental evidence
that some energy transfer processes can be usefully employed to
uplift the efficiency of the Yb anti-Stokes cooling process, supplying
an extra release of thermal energy.
The cooling performances of a YLF:5%Yb single crystal, where
a controlled Tm doping of 16 ppm (0.0016%) was added, were
investigated. Its performances were compared to single Yb-doped
YLF samples. A net efficiency increase, over the single 5% Yb doping,
was observed with the addition of a controlled Tm doping.
Figure 8.9 reports results of cooling measurements and a fit
model curve of the Yb–Tm co-doped YLF sample. Along with data
relative to the co-doped system are reported, for comparison,
cooling efficiency data relative to the barely Yb-doped 5% and 10%
YLF single crystals.
Yb
Yb
5%
%
10
5% Yb-Tm enriched
Figure 8.9 Experimental data points and fit model curve of Tm-enriched
YLF:5%Yb single crystal, along with cooling efficiency data
relative to single Yb doped at 5% and 10% YLF samples.
The fit of experimental data returned the following values for

the YLF:5%Yb-0.0016%Tm sample: ηext = 0.988 ± 0.001 and αb =
(1.9 ± 0.4) × 10–4 cm–1. For the cooling parameters relative to the
Yb-doped YLF single crystals we refer to Table 8.1. The addition
of controlled Tm doping in the 5% Yb-doped sample resulted in a
dramatic decrease of the background absorption parameter, with
respect to the single 5% Yb doping. Parasitic absorption and peak
efficiency values comparable to those obtained by increasing the
Yb doping level up to 10% were achieved via Yb–Tm co-doping. In
addition to that, the Yb–Tm co-doped sample, containing a lower
Yb concentration, exhibited a larger EQE with respect to the higher
Yb doping levels, which enabled a similar background absorption
decrease. Larger Yb concentrations uplift the absorption efficiency
but the strengthening of reabsorption phenomena tends to limit
the efficiency enhancement due to depleted fluorescence escape
rates. Virtuous energy transfer instead can improve the anti-Stokes
efficiency, supplying an extra release of thermal energy, but without
suffering from EQE reductions due to reabsorption-related effects.
A comparative spectroscopic analysis between single Yb-doped
and Yb–Tm co-doped samples was the starting point to investigate
the processes involved in the Yb–Tm energy transfer and their
contribution to the Yb anti-Stokes efficiency.
At first, the discrepancy in cooling performances was checked
to be unrelated to different absorption and emission properties of
Yb ions within Tm enriched and single Yb-doped samples. In Fig.
8.10 are shown the polarized absorption and fluorescence spectra,
acquired in the same experimental conditions, for the Tm-enriched
YLF:5%Yb and the single 5% Yb-doped YLF sample. The spectra
acquired did not exhibit any appreciable difference, within the
experimental uncertainty, in size or structure. Dynamic fluorescence
measurements returned similar values for the Yb transition lifetime
as well: τ = 2.1 ± 0.2 ms. The dissimilar cooling behavior could thus
be considered unrelated to different absorption and emission cross
sections of Yb ions.
However, visible fluorescence observed in laser cooling
experiments provided an indication that different energy transfer
processes were dominant in case of Yb–Tm co-doping. While single
Yb-doped YLF crystals showed a typical green emission when excited
in laser cooling experiments, the Yb–Tm co-doped sample exhibited
a characteristic blue fluorescence. To investigate processes involved
in the energy transfer from excited Yb ions, the emissions outside
the Yb region were observed in detail.
(a)
(b)
Figure 8.10 Polarized absorption (a) and fluorescence (a) spectra of the
Yb transition acquired, in the same experimental conditions,
for the Tm-enriched YLF:5%Yb sample (labeled as A) and the
single doped YLF:5%Yb (labeled as B). Fluorescence spectra
were acquired by exciting Yb ions at 940 nm along the c axis.
Visible and NIR fluorescence spectra were acquired, in the same

experimental conditions, for the Tm-enriched sample and the single
Yb-doped samples after laser excitation of the Yb ions at 940 nm.
Elemental analysis, performed by LA-IPC-MS [23], identified similar
amounts of other rare impurities. The discrepancies in emission
spectra between Tm-enriched and single Yb-doped samples could
thus be attributed exclusively to processes involved in the Yb–Tm
energy transfer. Figure 8.11 reports, for simplicity, only emission
spectra relative to Yb–Tm co-doped YLF and YLF:10%Yb samples,
which exhibit a comparable value of background absorption. All the
signals in the spectra are in arbitrary units, but the scale has been
respected, expected for the IR signal, because of a different detector.
14 YLF:5% Yb-0.0016% Tm 0.6
YLF:10% Yb YLF:Yb5%-Tm0.0016%
12 YLF:Yb10%
0.5
10
Intensity [a.u.]
0.4
8
20X 0.3
6
0.2
4
2 0.1
0 0.0
460 480 500 520 540 560 620 640 660 680
wavelength [nm] wavelength [nm]
4.5
YLF:Yb5%-Tm0.0016% 0.40
4.0 YLF:Yb10%
YLF:Yb5%-Tm0.0016%
0.35 YLF:Yb10%
3.5
0.30
3.0
Intensity [a.u.]
0.25
2.5
2.0 0.20
1.5 0.15
1.0 0.10
0.5 0.05
0.0 0.00
720 740 760 780 800 820 840 860 1800 1900 2000 2100 2200
wavelength [nm] wavelength [nm]
Figure 8.11 Emission spectra from the Tm-enriched YLF:5%Yb sample

(red data points) and YLF:10%Yb (blue data points) acquired,
in the same experimental conditions, after laser excitation of
the Yb transition at 940 nm along the c axis.
Although the structure and energy positions of the emission

bands were similar for all the samples, the Tm-enriched sample
clearly exhibited intense emissions bands at higher average energies
than that of excitation photons, which were strongly suppressed in
the YLF samples doped only with Yb ions. The blue emission, between
450 and 500 nm, is dominant along with intense NIR fluorescence,
between 750 and 850 nm, and red fluorescence between 630 and
680 nm.
The atomic-like structure of the emissions in Fig. 8.14 suggested
trivalent rare earth impurities as the sources of such bands. Test
samples of YLF single crystals, doped with other rare earth ions, and
tabulated energies of Stark sublevels [8] were employed to identify
the elemental sources of the observed fluorescence.
The emission bands in the blue, red, and NIR region were
identified with the Tm transitions 1G4 Æ 3H6, 3H4 Æ 3H6, and 1G4 Æ
3F respectively. The green and 2 µm bands, which exhibited similar
4,
intensity in the co-doped and single Yb-doped YLF samples, instead,
were identified with combined emissions from Er and Ho ions.
The internal transitions within Tm ions that give rise to blue, red,
and NIR fluorescence hence appeared those involved in a virtuous
energy transfer process, favorable for the anti-Stokes cooling
mechanisms.
In the Yb–Tm system, energy transfer typically results in blue
upconversion via nonresonant three-photon processes [1]. These
mechanisms involve three sequential steps of nonradiative transfers
of excitation from excited Yb ions to Tm, with strong heat release
via multiphonon emission. However, when the Yb concentration is
dominant over Tm ions, a cooperative sensitization process between
Yb ions becomes more likely than nonresonant three-photon
upconversion, due to the strengthening of the Yb–Yb interaction.
In the cooperative sensitization process a pair of excited Yb ions
couple and transfer their energy to one Tm ion, which gets excited
to the 1G4 manifold [16, 18]. Such a mechanism is peculiar of the
Yb–Tm system because the energy of the acceptor state (Tm, 1G4)
is about twice the energy of the donor excited state (Yb, 2F5/2) and
nonradiative quenching of the acceptor state is prevented by the
absence of underlying states close enough in energy.
To identify which mechanism was dominant in the Yb–Tm co-
doped sample, the fluorescence intensity at 477 nm was measured
as a function of the pump power at 940 nm. The fit of experimental
data returned a power law with an exponent value of 2.3 ± 0.2,
which indicates that the excitation of the 1G4 manifold was mostly
a two-photon process, that is, it essentially occurred via cooperative

sensitization.
The virtuous contribution of the Yb–Tm energy transfer to the
Yb anti-Stokes cooling mechanism was observed to be basically
related to the cooperative sensitization process. Indeed, while the
three-photon energy transfer involves strong phonon release and
thus detrimental heat generation, the cooperative sensitization
process is energetically favorable for heat removal. Via cooperative
sensitization of Yb pairs, the Tm 1G4 manifold is excited just above
the first Stark sublevel. Annihilation of lattice phonons is thus
required for thermalization of the electronic population within the
1G manifold. The following radiative emission from the 1G manifold
4 4
thus supplies an extra release of thermal energy from the lattice,
which usefully contributes to the anti-Stokes process in thermal
energy removal.
In addition to de-excitation of the 1G4 manifold to the ground
state, red fluorescence provides an additional channel of radiative
emission at higher energy than pumping photons, which does not
involve phonon release. It is worth to notice that the energy width of
the 1G4 manifold at 10 K is about 600 cm–1, comparable to that of the
Yb excited state (400 cm–1), and thus a similar amount of thermal
energy removal can be involved in both the Yb anti-Stokes cycle and
the Yb–Tm energy transfer.
In addition to that, several virtuous processes can be responsible
of the excitation of Tm ions to the 3H4 state. At first, radiative emissions
from the 1G4 manifold provide excitation of the 3H4 manifold without
heat dissipation. Energy transfer processes between excited Tm
ions and Yb ions in the ground state (Tm 1G4, Yb 2F7/2)Æ (Tm
3H , Yb 2F 3
6 5/2) can contribute to populate the H4 manifold with
simultaneous excitation of the Yb transition on the long-wavelength
tail of the absorption spectrum. Furthermore, endothermic Tm–
Tm upconversion (3F4, 3F4)Æ (3H6, 3H4) can likely contribute to
excitation of the 3H4 manifold, subtracting thermal energy from the
lattice. Due to the higher energy gap of the 3F4Æ3H4 transition with
respect to the 3F4Æ3H6 transition, indeed, such a process requires
phonons annihilation and thus thermal energy removal from the
lattice. The 10 K mismatch of about 1400 cm–1 requires absorption
of some phonons at room temperature. A relatively long lifetime (of
about 2 ms) was measured for the 3H4 manifold, which indicates
a negligible rate for the cross-relaxation process (3H6, 3H4)Æ (3F4,

3F ).
4
For specific relative concentrations of Yb donors and Tm
acceptors, several virtuous processes can be involved in the Yb–Tm
energy transfer, providing a positive contribution to the anti-Stokes
cooling process in heat removal.
Although further investigations are required in order to quantify
the efficiency of energy transfer processes as a function of decreasing
temperature, results achieved so far provide the first experimental
evidence that selected energy transfer processes can be employed to
enhance the efficiency of the anti-Stokes process. The potentialities
of this scheme still need to be fully explored. However, preliminary
results appear promising for further enhancement of the Yb
efficiency in fluoride crystals toward reaching temperatures close to
80 K, improving at the same time the cooling powers available at low
temperatures.
So far efficiency enhancements in Yb-doped fluorides have
been achieved, increasing the concentration of active ions. Such
an approach, however, is strongly limited because of reabsorption
phenomena, which become more efficient with the increase of the
Yb concentration. Controlled energy transfer processes provide,
instead, an efficient alternative that relies on supplying an additional
channel of thermal energy removal with efficiency only dependent
on the relative concentrations of donors and acceptors ions.
In addition to that, the Yb–Tm energy transfer could be usefully
employed to inhibit, a posteriori, energy transfer processes to
detrimental impurities, which involve heat generation in internal
processes or excitation transfer. Such a possibility reduces the
stringent constraints on the purity of starting materials, with huge
applicative effects. As mentioned earlier, purification of binary
fluorides below a certain grade is a quite difficult process, due to
the chemical inertness of these materials. The possibility to inhibit
a posteriori energy transfer to detrimental impurities would
provide an efficient tool to limit detrimental effects due to impurity-
mediated processes, hence imposing much less mandatory purity
requirements.
Investigations on the Yb–Tm energy transfer as a function of
temperature and relative concentrations of donors and acceptors
are currently underway in our laboratory in order to optimize the
References 269
efficiency of the energy transfer process and study its efficiency as

the temperature decreases.
8.5 Concluding Remarks

The efficiency of the optical cooling process has been investigated
in several Yb-doped YLF single crystals grown in our facility.
The increased Yb doping level resulted in significant efficiency
enhancement via diminishing background absorption. A virtuous
mechanism, based on energy transfer processes, has been identified,
providing the first experimental evidence of energy transfer–
assisted anti-Stokes cooling. Preliminary results achieved via Yb–Tm
co-doping in YLF crystals show a remarkable increase in cooling
performances with respect to the single Yb doping, promising for
enhancements of the cryogenic state-of-the-art performances
achieved in single Yb-doped fluorides. Such an approach possesses
huge potentialities toward intrinsic efficiency enhancements of
active materials, providing an extra release of thermal energy
via virtuous energy transfer from excited Yb ions. Moreover it
participates in inhibiting energy transfer processes to detrimental
impurities, reducing the effects of impurity-mediated effects on the
Yb anti-Stokes efficiency.
Acknowledgments
The authors would like to acknowledge I. Grassini for her competence
and care in preparing the samples. A. Volpi acknowledges support by
the European Space Agency under grant no. 4000108074/13/NL/
PA, “Cooling Effect on Fluoride Crystals.”
References
1. Auzel, F. (2004). Upconversion and anti-Stokes processes with f and d
ions in solids, Chem. Rev., 104, pp. 139–173.
2. De Loach, L. D., Payne, S. A., Chase, L. L., Smith, L. K., Kway, W. L., Krupke,
W. F. (1993). Evaluation of absorption and emission properties of Yb3+
doped crystals for laser applications, IEEE J. Quantum Electron., 29, pp.
1179–1191.
3. Di Lieto, A., Sottile, A., Volpi, A., Zhang, Z., Seletskiy, D. V., Tonelli, M.
(2014). Influence of other rare earth ions on the cooling efficiency in
Yb:YLF crystals, Opt. Express, 22, pp. 28572–28583.
4. Dong, G. Z., Zhang, X. L., Li., L. (2013). Energy transfer–enhanced laser
cooling in Ho3+ and Tm3+ co-doped lithium yttrium fluoride, J. Opt.
Soc. Am. B, 30, pp. 939–944.
5. Epstein, R. I., Buchwald, M., Edwards, B., Gosnell, T., Mungan, C. (1995).
Observation of laser-induced fluorescent cooling of a solid, Nature,
377, pp. 500–502.
6. Hehlen, M. P., Epstein, R. I., Inoue, H. (2007). Model of laser cooling
in the Yb3+ -doped fluorozirconate glass ZBLAN, Phys. Rev. B, 75, p.
144302.
7. Hehlen, M. P., Boncher, W. L., Melgaard, S. D., Blair, M. W., Jackson, R. A.,
Littleford, T. E., Love, S. P. (2014). Preparation of high-purity LiF, YF3
and YbF3 for laser refrigeration, Proc. SPIE, 9000, p. 900004.
8. Kaminski, A. A. (1996). Crystalline Lasers: Physical Processes and
Operating Schemes (CRC Press).
9. Kastler, A. (1950). Some suggestions concerning the production and
detection by optical means of inequalities in the populations of levels
of spatial quantization in atoms. Application to the Stern and Gerlach
and magnetic resonance experiments, J. Phys. Radium, 11, pp. 255–
265.
10. Kuhn, H., Fredrich-Thornton, S. T., Krankel, C., Peters, R., Petermann, K.
(2007). Model for calculation of radiation trapping and description of
the pin-hole method, Opt. Lett., 32, pp. 1908–1910.
11. Melgaard, S. D. (2013). Cryogenic Optical Refrigeration: Laser Cooling
of Solids Below 123K, PhD dissertation, University of New Mexico,
Albuquerque, NM.
12. Melgaard, S. D., Albrecht, A., Hehlen, M. P., Seletskiy, D. V., Sheik-Bahae,
M. (2014). Optical refrigeration cools below 100K CLEO 2014, OSA
Tech. Dig., paper FTh4D.4.
13. Melgaard, S. D., Seletskiy, D. V., Di Lieto, A., Tonelli, M., Sheik-Bahae,
M. (2013). Optical refrigeration to 119K, below National Institute of
1588–1590.
14. Melgaard, S. D., Seletskiy, D. V., Polyak, V., Asmerom, Y., Sheik-Bahae, M.
(2014). Identification of parasitic losses in Yb:YLF and prospects for
optical refrigeration down to 80K, Opt. Express, 22, pp.7756–7764.
References 271
15. Nemova, G., Kashyap, R. (2011). Alternative technique for laser cooling
with super irradiance, Phys. Rev. A, 83, p. 013404.
16. Ovsyakin, V. V., Feofilov, P. P. (1966). Cooperative sensitization of
luminescence in crystal activated with rare earth ions, Sov. Phys. JETP
Lett., 11, pp. 317–318.
17. Payne, S. A., Chase, L. L., Smith, L. K., Kway, W. L., Krupke W. F. (1992).
Infrared cross-section measurements for crystal doped with Er3+,
Tm3+ and Ho3+, IEEE J. Quantum Electron., 28, pp. 2619–2630.
18. Pei., X., Hou, Y., Zhao, S., Xu, Z., Teng, F. (2005). Frequency upconversion
of Tm3+ and Yb3+ codoped LiYF4 synthesized by hydrothermal
method, Mater. Chem. Phys., 90, pp. 270–274.
19. Ruan, X. L., Kaviany, M. (2006). Enhanced laser cooling of rare earth ion
doped nanocrystalline powders, Phys. Rev. B, 73, p. 155422.
20. Seletskiy, D. V., Hehlen, M. P., Epstein, R., Sheik-Bahae, M. (2012).
Cryogenic optical refrigeration, Adv. Opt. Photonics, 4, pp. 78–107.
21. Seletskiy, D. V., Melgaard, S. D., Bigotta, S., Di Lieto, A., Tonelli, M., Sheik
23. Tiepolo, B., Bottazzi, P., Palenzona, M., Vannucci, R. (2003). A laser probe
coupled with ICP: double focusing sector; field mass spectrometer
in situ analysis of geological samples and U-Pb dating of zircon, Can.
Mineral., 41, pp. 259–272.
24. Volpi, A., Di Lieto, A., Tonelli, M. (2015). Novel approach for solid state
cryocoolers, Opt. Express, 23, pp. 8216–8226.
Chapter 9
Interferometric Measurement of Laser-

Induced Temperature Changes
B. Rami Reddy
Alabama A&M University, Department of Physics, Normal,
AL 35762, USA
rami.bommareddi@aamu.edu
9.1 Historical Development of Temperature

Sensors
Temperature is the most important parameter in scientific and
industrial environments. The development of temperature sensors
is continually evolving for more than two centuries. Any material
whose property changes quickly with temperature and is reversible,
can be used as a thermometer. This property can be pressure, volume,
length, refractive index, color, etc. The efforts of Fahrenheit and later
Celsius led to the development of the mercury-in-glass thermometer.
Modern infrared (IR) thermometers scan the forehead quickly
to reveal the body temperature. Some of the popular devices are
thermocouples, resistive temperature devices (RTDs), thermistors,

www.panstanford.com
274 Interferometric Measurement of Laser-Induced Temperature Changes
thermostats, silicon-based sensors, and IR pyrometers [1]. Each

type of device operates in a specific temperature range. They are
installed in consumer appliances, process industries, and aircrafts.
Thermocouples are cheap and rugged, some operate up to 1800°C,
and they are installed in furnaces, combustion engines, chemical
reactors, and many others. Silicon sensors operate up to 150°C and
are used in electronic circuit boards. RTDs are wire windings of
copper or another metal, whose resistance increases linearly with
temperature, and they are useful for temperature measurement
up to a few hundred degrees. Conventional thermometers are
inexpensive, but they have limited applications.
Though several different types of temperature probes are
available there is no ideal device that works in all types of
environments, for temperature measurement. Optical materials are
either heated up or cooled down when a laser beam passes through
them [2, 3]. Some materials require longer times to reach thermal
equilibrium. Contact thermometers are not suitable for temperature
measurement along the beam path. These measurements require
noncontact sensing devices, which are preferred in many cases. To
address these specific needs optical techniques are investigated
for temperature measurement, using rare earth ion–doped solids.
These are alternate techniques for temperature measurement when
conventional techniques are not suitable.
9.1.1 Optical Sensor Technologies

IR thermometers and imagers are noncontact devices that use
IR radiation to infer the temperature of an object. With the
advent of lasers a range of optical sensors are developed. Optical
measurements involve the detection of light. Any parameter that
changes with temperature can be used to estimate the sample
temperature, such as changes in emission intensity, lifetime, optical
path length, or phase. Optical sensors can be configured as point
sensors or distributed temperature sensing (DTS) systems and
the latter uses mainly optical fibers [4]. Intensity measurement
sensors are based on Raman scattering [5], fluorescence [6], or
spectral transmission changes [7]. As the temperature of a material
increases, higher vibrational levels of molecules gain population
due to Boltzmann distribution. Raman scattering methods took
Historical Development of Temperature Sensors 275
advantage of this phenomenon to estimate the temperature in the

combustion studies [5]. Raman scattering is the principle behind the
operation of distributed temperature sensor. If a fiber’s temperature
changes at one or more points along the length of a fiber, thermal
effects induce lattice vibrations locally, and a propagating pulsed
laser beam undergoes Raman scattering at those points. Time taken
by the backscattered Raman signals to arrive at the input end of
the fiber is measured by the optical time domain reflection (OTDR)
technique to locate the heated spots. The intensity of the anti-Stokes
line is temperature dependent and the ratio of anti-Stokes to Stokes
line intensities is used to derive the fiber temperature. Commercial
devices measure temperatures up to 700°C with a linear resolution
of 1 m. A fiber Bragg grating (FBG) is produced by creating a periodic
variation in the refractive index in the fiber core, which reflects a
specific wavelength. The reflected light wavelength depends on the
grating period and the latter depends on the grating temperature.
This principle was used to design FBG temperature sensors for
several applications [8], including laser cooling studies [9].
9.1.2 Transmission Spectral Measurements for High-

Temperature Measurement
The device consists of an optical fiber made of sapphire, Y2O3, or
another material [7]. One end of the fiber is coated with a rare-
earth-ion-doped film that is in contact with the surface under
measurement. The light output of the fiber passes through a band-
pass filter (or spectrometer) before being detected, whose intensity
is related to the sample temperature. This device is useful for
temperature measurement from 400°C to 2000°C and operates even
in reactive and corrosive environments. It has a fast response time of
1 ms and a temperature resolution of 1°C.
9.1.3 Interferometric Sensors

Interferometric sensors are demonstrated using several different
techniques and configurations [10–14]. In the prelaser era
interferometers utilized either mercury or sodium light sources
whose coherence lengths are very small. So the commercial products
were mainly centered on testing optical surfaces or components
[15]. After the advent of lasers the applications of interferometry

grew exponentially over time, which was attributed to the long
coherence lengths of the laser beams [16]. Two coherent beams can
be derived from a laser source by the division of wavefront or division
of amplitude, which are suitable for the design of several kinds of
interferometers. Interferometers can be classified as two-beam
or multiple-beam devices. The fringe intensity varies sinusoidally
in a two-beam interferometer. In a multiple-beam interferometer,
each pair of beams contributes a Fourier component to the fringe
pattern. Interferometers are classified based on the method used to
derive the two beams. If the radiation coming from a point source
passes through two or more apertures, it is said to be separated
into beams by division of wavefront; they are (a) Young’s double-slit
experiment, (b) Lloyd’s mirror, and (c) Rayleigh interferometer [14].
Most of the interferometers derive the two beams by the division of
amplitude. Interferometers based on division of amplitude are the
Mach–Zehnder interferometer, Michelson interferometer, Sagnac
interferometer, and lateral shearing interferometer [10, 14]. For
many applications, it is desirable to have an optical arrangement
in which the two interfering beams travel along separate paths
before they are recombined. This led to the development of a large
number of interferometers for specific purposes. A polarization
interferometer results when light is separated into ordinary and
extraordinary rays (by using a calcite prism, Wollaston prism,
Rochon prism, or Nicol prism). The Mach–Zehnder interferometer
can be set up in a large space to accommodate big chambers in its
path length for in situ monitoring. Alignment of the optics is relatively
simpler if the mirrors are replaced by pentaprisms [17]. To facilitate
a variable path difference between the two beams, the mirrors are
replaced by corner cubes in the Michelson interferometer, and the
reflecting mirrors are replaced by mirror pairs in the Mach–Zehnder
interferometer. Such a modified version was used in the design of
a wavemeter [18, 19]. Straight-line fringes of equal thickness are
produced by the Michelson interferometer whenever a wedge-
shaped air gap is introduced between one mirror and the image of
the other in the former. A Sagnac interferometer is extremely stable
because the two beams travel in opposite directions in a closed path.
The beam splitters and mirrors of a Mach–Zehnder interferometer
are adjusted so that the optical beams are approximately parallel
Rare Earth Luminescence Spectroscopy and Temperature Sensors 277
and the paths travelled by both the beams are equal; the beams
form a rectangle or a parallelogram. Thus a collimated input beam
generates two wavefronts. If the wavefronts make a small angle with
each other a nonlocalized interference pattern is seen, consisting of
equispaced straight fringes parallel to the line of intersection of the
wavefronts. If there is an angle between the beams after exiting the
second beam splitter then localized fringes are formed. On the other
hand if there is no angle between the beams then nonlocalized fringes
are formed. The Mach–Zehnder interferometer is a much more
versatile instrument than the Michelson interferometer because
the optical beams are widely separated [20–23]. The beams travel
the path only once and the fringes can be localized at any desired
point. Because of this, it has been extensively used in studies of gas
flow, combustion, plasma density, and diffusion, where changes in
refractive index occur, which can be related to changes in pressure,
temperature, or the relative concentrations of different components
of a mixture [13, 23]. However, adjustment of the interferometer
is tedious to get fringes of good visibility, since a displacement of a
mirror results in a shift of the plane of localization as well as a change
in the optical path difference (OPD). The need for this procedure can
be eliminated by using pentaprisms in place of mirrors.
9.2 Rare Earth Luminescence Spectroscopy and

Temperature Sensors
Rare earth ion–doped solids are extensively investigated because
of their applications for the development of lasers, luminescent
devices, optical memories, and optical refrigerators [24–26]. For a
clear understanding of laser cooling/heating a brief description of
rare earth spectroscopy is given here. In general, rare earths form
trivalent compounds. Rare earth ions have an unfilled 4f shell whose
active electrons are shielded from the crystal field by the 5s and
5p orbitals. Consequently the crystal field splitting is smaller than
the spin–orbit splitting. As a result the center of gravity of each
multiplet approximately remains the same as that of the free ion but
exhibits slight changes from one host to the other. In the gas phase,
transitions among the levels of 4fn configuration are forbidden.
However, when the ions are embedded in a solid medium they are
weakly allowed due to admixing of 4fn wave functions with those of

4fn–15d1 wave functions at noncentro symmetric sites. Excited levels
have relatively long lifetimes and group theory predicts allowed
transitions for ΔJ ≤ 6 [27]. Excited ions in a solid relax radiatively
and nonradiatively [28], which are represented by a line arrow and
a wavy arrow in Fig. 9.1. The nonradiative relaxation depends on
the guest–host combination. For a given material the nonradiative
relaxation depends on the energy gap between the excited level and
its lower level, the cutoff phonon frequency of the host material, and
the phonon density. The cutoff phonon frequency depends on the
reduced mass of the molecule and force constants. For example, the
cutoff phonon frequencies in LaF3, LaCl3, and LaBr3 are 350, 260,
and 175 cm–1, respectively [29]. When the energy gap ΔE32 between
levels 3 and 2 is bridged by one or two phonons the upper level
relaxes nonradiatively. If the energy gap ΔE10 between levels 1 and
0 requires eight or more phonons, nonradiative relaxation does not
occur between levels 1 and 0. For intermediate energy gaps as ΔE21
between levels 2 and 1 both radiative and nonradiative relaxation
phenomena occur simultaneously. For intermediate energy gaps
the higher levels relax radiatively and nonradiatively. Therefore,
the measured lifetime τ of the excited level i is related to sum of the
relaxation rates as [28]
τ–1(i) = Σijωr + Σijωnr (9.1)
where Σωr is the total radiative relaxation rate to all the lower levels
and Σωnr is the total nonradiative relaxation rate to all the lower
levels, represented by j. The nonradiative relaxation rate increases
at higher temperatures and hence the lifetime of the excited level
decreases at higher temperatures [28]. Based on this phenomenon
fluorescence lifetime sensors were designed to measure the sample
temperature [30]. However, this technique is not relevant for the laser
cooling studies because the nonradiative relaxation rate has to be
negligible for laser cooling of solids. However, fluorescence intensity
ratio based sensors are useful for temperature measurement if the
energy gap between levels 2 and 3 is comparable to thermal energy.
In general, emission occurs from level 2 and a weak emission
from level 3. As the sample temperature increases, the population
in level 3 increases due to thermalization and the intensity of λ30
increases at higher temperatures. This is true for Yb3+-doped solids.
The population Ni (i = 0 to 3) of an energy level is dependent on

temperature and the population ratio is given by
N2 g2 -DE20 /kT
= e (9.2)
N0 g0
N1 g1 -DE10 /kT
= e (9.3)
N0 g0
3
2
l21
1
Laser
l10 l20 l30
0
Figure 9.1 Energy levels and relaxation phenomena in a rare earth ion–
doped solid. Radiative and nonradiative relaxation phenomena
are represented by downward straight-line arrows and wavy
arrows. Emission from level 3 (dotted line) occurs if its
population is sustained by thermalization.
After taking a ratio of the above equations, it yields

N2 g2 -DE21 /kT
= e (9.4)
N1 g1
where gi is the degeneracy of level i and ΔEij is the energy gap between
levels i and j, T is the sample temperature, and k is Boltzmann’s
constant. The transition intensity, I, between levels 1 and 0 is given
by
I10 µ N1v10A10 (9.5)
where ν10 is the frequency corresponding to the wavelength λ10 and
A is the transition probability. The intensity ratio, R, of the transitions
for 2 Æ 0 and 1 Æ 0 gives
DE21 DE21
I20 N2n20 A20 n20 A20 g2 - -
R= = = e kT = (const)ee kT (9.6)
I10 N1n10 A10 n10 A10 g1
The above equation can be rearranged as

DE21 b
lnR = a - =a- (9.7)
kT T
where a and b are constants. A plot of logarithmic R versus T
increases linearly with temperature. A plot of ln R versus 1/T
decreases linearly with temperature. At lower temperatures the
ratio R is smaller. However, this method requires the development
of a calibration plot by measuring the intensity ratio as a function
of sample temperature. An unknown sample temperature can be
derived by utilizing the calibration plot and from a measurement of
the fluorescence intensity ratio. Such a method was demonstrated
for rare earth–doped samples for high-temperature measurements
[31]. This technique should work equally well even if the sample
temperature decreases. For low-temperature measurement the
above equation predicts sample temperature accurately if the peaks’
intensities are measured accurately. Because of these reasons, a
linear variation was observed in a limited temperature range, up to
140°C in an Er3+-doped silica waveguide [32] and up to 96°C in an
Er3+-doped silica fiber [33], though mK sensitivity was reported in
the past [34].
9.2.1 Differential Luminescence Measurement

The differential luminescence technique works on similar princi-
ples. Populations among different energy levels, for example, levels
2 and 3 of Fig. 9.1, change whenever the sample temperature chang-
es. Consequently spectral features, relative intensities of different
peaks, linewidths, and in some cases the fluorescence band shape
change significantly. In this case a calibration plot needs to be devel-
oped in advance by cooling the sample to different temperatures and
acquiring the spectra with a charge-coupled device (CCD) camera.
Spectral subtraction reveals whether the material is heating or cool-
ing. The technique reveals very small changes in temperature [35].
9.2.2 Yb3+-Doped YAG Spectra

The transmission spectrum of Yb3+-doped YAG was obtained by
exciting the sample with a global lamp source and collecting the
sample transmission with a low-resolution spectrometer. The
resulting spectrum was mathematically manipulated to depict the

absorption spectrum (Fig. 9.2). It reveals a strong absorption at 940
nm and a weaker absorption at 1030 nm. The absorption strength
at 915 nm is weaker than that at 940 nm. The laser wavelength 915
nm excites the higher stark levels of the 2F5/2 multiplet (Fig. 9.4).
Excited-state ions relax nonradiatively (multiphonon relaxation) to
the lowest Stark level of the excited state that emits at 1030 nm. The
laser-induced heating occurs mainly along the path of the laser beam.
In such cases, indirect methods are used to estimate the sample
temperature accurately. One such method is Raman scattering, in
which the peak intensities depend on the populations of the levels
involved, which is a function of the sample temperature. Here we are
proposing interferometric methods to measure heating or cooling
along the beam path in real time.
Figure 9.2 Absorption spectrum of YAG:Yb3+ [36].
Figure 9.3 Fluorescence observed under 915 nm laser excitation.

10 200 cm-1
2F
5/2
9 800
700
2F
7/2
0
Figure 9.4 Partial energy level diagram of YAG:Yb3+. Upward and
downward arrows represent laser excitation and fluorescence.
Wavy lines represent nonradiative relaxation.
9.3 Mach–Zehnder Interferometer

A Mach–Zehnder interferometer was set up with a 514.5 nm green
laser beam to measure temperature changes in Yb3+-doped YAG
induced by another laser [36]. The 514.5 nm green laser beam
is separated into two parts by a beam splitter, BS1, and the two
beams are arranged to form a rectangle (Fig. 9.5). Beam 1 passes
through the sample. Beams 1 and 2 are recombined by the beam
splitter BS2. Mirrors M1 and M2 are adjusted so that a wedge
forms between the two wavefronts. As a result linear fringes are
produced, which are projected on a screen by a lens (Fig. 9.6). A
detector having a pinhole opening monitors the intensity of a fringe.
A Ti:sapphire laser beam (915 nm) is launched into the sample
that propagates in a direction opposite to that of the green laser
beam. Both the beams are collinear and counterpropagating. The
915 nm laser excites Yb3+ ions from the ground state to the excited
state, 2F5/2 (Fig. 9.4). Excited-state ions exhibit strong emission at
1030 nm (Fig. 9.3). During radiative relaxation some ions end up in
higher Stark components of the ground state, 2F7/2 (Fig. 9.4). The
Mach–Zehnder Interferometer 283
nonradiative relaxation processes in the ground and excited states

heat the material along the beam path. Surface-mounted contact
thermometers may not reveal the sample temperature accurately
if the material conductivity is small. The optical path length of the
beam is dependent on sample temperature because the refractive
index and sample length are dependent on sample temperature.
In the interferometric arrangement the fringes appear to move
when the sample temperature is continuously changed. As a result
intensity of the fringe pattern varied continuously with time, and
the detector output followed such a variation (Fig. 9.7). Material
heating due to nonradiative relaxation occurs instantaneously
when it is exposed to a 915 nm laser beam. Therefore, the period
of oscillation is small in the beginning. The period of oscillation
increases with time because the crystal temperature is slowly
approaching the steady-state value. Therefore, the oscillatory output
of the detector exhibits chirping [36]. Once it reaches steady-state
value the material temperature does not change anymore and hence
the oscillation stops. The detector output is constant if the material
does not absorb the probe laser beam. For 300 mW of a 915 nm laser
beam, the sample temperature increased 7.1 ± 0.8°C [31]. A similar
situation occurs when the heating laser beam was blocked (Fig. 9.8).
detector
screen
heat cool
computer
514.5 nm
BS2
M1 915 nm
YAG: Yb3+
BS1 M2
2
514.5 nm
Figure 9.5 A 915 nm laser beam heats the YAG:Yb3+ sample. A 514.5 nm
beam was used to setup a Mach–Zehnder interferometer to
monitor temperature changes [36].
Figure 9.6 Interferometer fringes produced by a 514.5 nm laser. The dark

spot is the pinhole in front of the detector [36].
Time (s)
Figure 9.7 Temporal variation of the detector output during laser

Ti:sapphire heating.
9.3.1 Theory of Mach–Zehnder Interferometer

The laser beam passing through the sample changes its temperature.
As the sample temperature is varied its refractive index changes. So
the effective optical path length ηL of beam 1 changes continuously
and this change causes the fringes to move. The apparent speed of
the fringes depends on the rate of change dη/dt and hence on dT/
dt. The path lengths of beams 1 and 2 are η1(L2 + L1 – L3) + η2L3 and
η1(L1 + L2), where η1 and η2 are refractive indices of air (η1 = 1.0)
and YAG (η2 = 1.823), respectively; L1 and L2 are the dimensions of
the rectangle; and L3 is the sample length. The OPD between the two
beams 1 and 2 is given by OPD = (η2 – η1)L3. Here, η2 and L3 are
dependent on sample temperature.
The refractive index and sample length are temperature
dependent, which are described by
∂h
h3 = h0+ DT (9.8)
∂T
∂L
L3 = L0+ DT (9.9)
∂T
where η0 and L0 are the room-temperature values of refractive index
and sample length, respectively. When the Ti:sapphire laser beam
passes through the sample its temperature changes from T1 to T2.
However, the change in OPD is given by
Δ(OPD) = Δ(η2L3 – η1L3) (9.10)
Since η1 = 1.0 for air,
Δ(OPD) = Δ(η2L3) – Δ(η1L3) = Δη2L3 + η2ΔL3 – ΔL3 (9.11)
∂h ∂L ∂L
D(OPD) = L3 DT + h3 DT - DT (9.12)
∂T ∂T ∂T
Hereafter L3 = L0 = L (sample length) and η2 = η0 = η (sample
refractive index).
∂h ∂L ∂L
D(OPD) = L DT + h DT - DT (9.13)
∂T ∂T ∂T
Ê 1 ∂h 1 ∂L 1 ∂L ˆ
D(OPD) = Lh Á DT + DT - DT ˜ (9.14)
Ë h ∂T L ∂T Lh ∂T ¯
Equation 9.14 can be written as
Ê 1 ∂h aˆ
D(OPD) = Lh Á + a - ˜ DT (9.15)
Ë h ∂T h¯
where α is the linear thermal expansion coefficient of the material,
given by
1 ∂L
a= (9.16)
L ∂T
If N fringes translated for a given temperature change ΔT, then

Δ(OPD) = λN. Equation 9.15 reduces to
Ê aˆ
l N = Lh Á g - ˜ DT (9.17)
Ë h¯
where
1 ∂h
g = +a (9.18)
h ∂T
The temperature variation of the sample along the beam path is

given by
lN
DT = (9.19)
Ê aˆ
Lh Á g - ˜
Ë h¯
From the above equation it is clear that ΔT is a function of N, L, η,
α, and γ. So the accuracy of the ΔT estimate depends on the accuracy
with which these parameters were measured. The uncertainty in
temperature change ΔTu is calculated by the method of standard
deviations, as follows.
DTu
2 2 2 2 2
Ê ∂T ˆ Ê ∂T ˆ Ê ∂T ˆ Ê ∂T ˆ Ê ∂T ˆ
= Á ( DNu )2 + Á ( DLu )2 + Á ˜ ( Dhu )2 + Á ( Da u )2 + Á ( Dg u )2
Ë ∂N ˜¯ Ë ∂L ˜¯ Ë ∂h ¯ Ë ∂a ˜¯ Ë ∂g ˜¯
(9.20)
where the partial derivatives are obtained by differentiating Eq. 9.19
with respect to each parameter as
∂T l 1
=- (9.21)
∂N Lh Ê aˆ
ÁË g - h ˜¯
∂T l N 1
=- (9.22)
∂L hÊ a ˆ L2
ÁË g - h ˜¯
∂T -gl N 1
= (9.23)
∂h L Ê aˆ h
2 2
ÁË g - h ˜¯
∂T l N 1
= (9.24)
∂a L Ê aˆ h
2 2
ÁË g - h ˜¯
∂T l N h
= (9.25)
∂g L Ê aˆ h
2 2
ÁË g - h ˜¯
The errors are estimated by substituting the respective values

into Eqs. 9.20 to 9.25.
As soon as the Ti:sapphire laser beam was turned on the fringes
appeared to move and the detector output was oscillatory. For a 300
mW laser beam it took about 2000 s to reach the steady-state value
and the detector output revealed three fringes in this time interval
(Fig. 9.7). By substituting the values for the YAG:Yb3+ sample length
L = 0.015 m, the number of fringes N = 3, refractive index η = 1.823
[37], coefficient of thermal expansion α = 7.8 × 10–6/K, fractional
change in optical path length at 310 K is γ = 1.1 × 10–5/K [38, 39],
and λ = 514.5 nm in Eq. 9.19, the change in sample temperature is
estimated as 7.1±0.8°C. The reliability of the temperature estimate
depends on the accuracy [31] of the other parameters α, β, and γ. The
sample absorption at 915 nm is small and hence the temperature
change of the material is small. If the Ti:sapphire laser is turned
off the crystal should cool down to its steady-state value. One such
curve observed is shown in Fig. 9.8. The crystal temperature almost
reached the steady-state value in about 900 s. However, the time
taken to reach the steady-state value during the crystal cooling is
smaller than that observed during crystal heating. The direction of
motion of the fringes during crystal cooling should be opposite in the
direction to that of heating. To reveal this feature another curve was
recorded, which is shown in Fig. 9.9. When the oscillatory detector
output was decreasing, the pump laser beam was blocked manually.
The detector output starts increasing (phase reversal), which
confirms that the material is cooling. From 0 to 1000 s (Fig. 9.9) the
crystal was heated by the Ti:sapphire laser beam, which was blocked
after 1000 s. The detector output was on the trailing side of the peak
value at time, t = 1000 s, and the fringe period was relatively longer
between 500 to 1000 s. When the Ti:sapphire laser was blocked the
material cools faster and hence the fringe period was smaller from
1000 to 1200 s [36]. As the crystal temperature was reaching the

steady-state value the fringe period increased again between 1200
and 2000 s. These results indicate that the crystal temperature does
not reach the steady-state value instantaneously. The required time
depends on the crystal parameters like thermal conductivity. If
there is any residual absorption at 514.5 nm due to the presence of
trace amounts of an unknown impurity, that also will contribute to
the material heating. At time t = 0, whether the signal rises or falls
depends on the proximity of the detector pinhole with respect to the
fringe and the direction of fringe movement.
Time (s)
Figure 9.8 Temporal variation of the detector output during cooling (915
nm laser was turned off at t = 0 s).
10.5
Laser off (cooling)
Intensity (a.u.)
8.5
6.5
Laser on (heating)
4.5
0 500 1000 1500 2000
time (s)
Figure 9.9 Temporal evolution of the detector output during the time of
laser heating (0 to 1000 s) and laser off/cooling (1000 to 2000
s).
Though we have demonstrated the application of the Mach–

Zehnder interferometer for laser heating the same configuration
Optical Heterodyne Technique 289
works equally well to investigate laser cooling of solids. Of course,

the difference is in the direction of fringe motion, as depicted in Fig.
9.9. The only stringent requirement is that the heating (or cooling)
laser and the probe laser beams overlap in the crystal.
9.4 Optical Heterodyne Technique

In general, a typical interferometer involves two beams, of which
one is a reference beam and the other is a measurement beam. The
measurement beam either gets reflected or passes through the
material under investigation. Interferometry deals with either fringe
counting or phase measurement [10, 14–16]. In a fringe-counting
experiment an ordinary He–Ne or another low-power laser of long
coherence length is required. For phase measurements a stabilized
two-frequency Zeeman laser is required [40]. If a Zeeman laser is
not available, the required two frequencies can be produced with a
modulator [41].
9.4.1 Principle of the Optical Heterodyne Detection

Technique
Optical frequencies are of the order of 6 × 1014 Hz or larger. When
such a wave falls on a detector, it does not respond to such fast
variations. So, the detector output is a DC current whose amplitude
is proportional to the intensity of light (Fig. 9.10). The two Zeeman
laser frequencies ν1 and ν2 are slightly different, and their difference
Δν = ν2 – ν1, is in the radiofrequency (RF) range. When two such
optical waves, ν1 and ν2, fall on a detector, its output is a DC current
that is modulated by the RF current at Δν. The frequency of the RF
component equals the difference between the two optical frequencies
(1.5 MHz for HP 5501A laser and 250 kHz for OPTRALITE laser). If
the two beams travel slightly different path lengths and then fall on
a detector, the resultant amplitude is given by [40]
E = E1 + E2 = Aei(k1z1 – ω1t) + Aei(k2z2 – ω2t), (9.26)
where A, ωi, ki, and zi (i = 1, 2) are, respectively, the amplitude, angular
frequency, wavenumber, and path length of each wave. Detector
output current, I, is proportional to the intensity of the beams,
I = EE* = I0 + IRF (9.27)
where
IRF = I0 cos[2πΔνt + ΔΘ(z)] (9.28)
where ΔΘ(z) = Θ1(z1) – Θ2(z2). Θ1(z1) is the phase of the reference
beam and Θ2(z2) is the phase of the measurement beam. As the
material is cooled (or heated) the optical path length travelled by
the measurement beam changes and, hence, its phase, Θ2(z2), also
changes. Consequently the phase difference between the two waves,
ΔΘ(z) changes continuously so long as the material temperature
changes. ΔΘ(z) is related to the path length change Δz as
ΔΘ(z) = (2π/λ)Δz (9.29)
where λ is the wavelength corresponding to the center frequency of
the Zeeman laser.
6 ×1014 Hz
ν1 dc current
ν1
ν2 dc current + rf frequency
rf frequency = ν2 - ν1 = 250 kHz
Figure 9.10 Conceptual diagram of optical heterodyne detection. The

rectangle represents the detector.
ΔΘ is measured by a phase meter. The OPTRALITE laser head

provides the 250 kHz RF for reference purposes. For an HP laser the
reference beam has to be derived from a second detector output.
The reference and measurement RFs (detector outputs of equal
frequency) are given to the two inputs of an HP 3575A phase meter,
whose output (in DC volts) is proportional to the phase difference
between the two RF inputs. The optical heterodyne technique was
applied in the past for accurate measurement of surface roughness
of coated mirrors [40, 42] and crystal growth rate measurement
[43].
9.4.2 Application of Optical Heterodyne Detection

Technique to Laser Cooling/Heating of Solids
Figure 9.11 shows the experimental configuration of a Mach–
Zehnder interferometer configuration, required to implement optical
heterodyne detection scheme for laser cooling or heating of solids.
A milliwatt Zeeman laser produces two collinear and orthogonally

polarized beams, whose frequencies differ by 1.5 MHz for an HP
laser and 250 kHz for an OPTRALITE laser. For some applications, a
setup similar to that of a Mach–Zehnder interferometer is preferred
because the measurement beam passes through the sample once.
The p- (|) and s- (•) polarized beams are spatially separated by a
polarizing beam splitter (PBS). Based on the optical configuration
the mathematical model has to be developed accordingly. For
the measurement of temperature changes in the experiments of
laser cooling or heating of solids, the heterodyne technique works
equally well. Whenever a material is heated or cooled its length
as well as refractive index change. A measurement of optical path
length variation of a beam, which passes through the material,
reveals the sample temperature change. So, prior knowledge of
the temperature dependence of the material’s refractive index and
thermal expansion coefficient is needed. For the implementation of
the optical heterodyne technique a Mach–Zehnder interferometer
configuration (Fig. 9.11) is a straightforward approach whose
mathematical description is discussed below. The two orthogonally
polarized beams of the Zeeman laser are spatially separated by a PBS,
of which one is a reference beam (ν1) and the other is a measurement
beam (ν2) that passes through the material under investigation. In
this case, the material is heated or cooled by another laser beam that
gets reflected by the dichroic mirror (DM), which is collinear with
the probing beam but passes in a direction opposite to the probe
beam (Fig. 9.11). The laser beam either heats up or cools down the
material, depending on the sample and beam wavelength. Say ν1 and
ν2 travel approximately equal distances, and their OPD is within the
coherence length of the laser. The refractive indexes of air and the
material are represented by η1 and η2, respectively. L1 and L2 are
the lengths of the interferometer arms and L3 is the sample length.
D1 and D2 represent the optical path lengths of the reference and
measurement beams, respectively, which are given by [44]
D1 = η1(L1 + L2) (9.30)
D2 = η1L2 + η1(L1 – L3) + η2L3 (9.31)

The OPD is given by
OPD = D2 – D1 = η2L3 – η1L3 (9.32)
If we assume that the refractive index of air, η1 = 1, then

OPD = η2L3 – L3 (9.33)
PBS
L1 M1
V1
Zeeman Reference
Computer
laser (rf)
L2
V2
h2 detector
DM
M2
rf
L3 POL Phase meter
Ti:Sapphire laser
Figure 9.11 Optical configuration of the heterodyne detection scheme.
Hereafter, we drop the subscripts and use η for the refractive

index of the material and L for sample length
OPD = ηL – L (9.34)
As the sample temperature changes η and L change accordingly.
Therefore a change in the OPD is given by
Δ(OPD) = Δ(ηL) – ΔL = ΔηL + ηΔL – ΔL (9.35)
The refractive index and sample length are temperature
dependent. Equation 9.35 is similar to Eq. 9.11 and further
simplification will reduce this equation to
∂h ∂L ∂L
D (OPD) = DTL + h DT - DT (9.36)
∂T ∂T ∂T
Ê 1 ∂h 1 ∂L 1 ∂L ˆ
D (OPD) = hL Á + - DT (9.37)
Ë h ∂T L ∂T hL ∂T ˜¯
where α and γ have their usual meaning. The change in the OPD is
simplified as
Ê aˆ
D(OPD) = hL Á g - ˜ DT (9.38)
Ë h¯
The OPD is related to the phase difference between the two
beams as
2p 2p Ê aˆ
DQ = D(OPD) = hL Á g - ˜ DT (9.39)
l l Ë h¯
By rearranging the above equation, the temperature difference is
estimated as [44]
lDQ
DT = (9.40)
Ê aˆ
2phL Á g - ˜
Ë h¯
The probe laser wavelength is 632.8 nm. The values of α and

g are estimated from the sample data [38, 39] using Eqs. 9.14 and
9.15. ΔΘ can be measured to an accuracy of 0.1° with a phase meter
(resolution of Model HP 3575A phase meter). For YAG material α = 7.8
× 10–6, g = 11 × 10–6, L = 0.0105 m, and η = 1.823. If ΔΘ is 0.1°, which
is the phase meter resolution, then Eq. 9.40 predicts a temperature
change of 1.4 mK (theoretical limit). This accuracy is comparable
to that obtained with the fluorescence ratio method [34] and the
differential luminescence measurement method [35]. However, the
measurement of such a small change in temperature may not be
possible because the accuracy is limited by the uncertainty, which is
calculated by the propagation of errors [45] as
DTu
2 2 2 2 2 2
Ê ∂T ˆ Ê ∂T ˆ Ê ∂T ˆ Ê ∂T ˆ Ê ∂T ˆ Ê ∂T ˆ
= Á ˜ ( Dq u )2 + Á ( DLu )2 + Á ( Dhu )2 + Á ( Dg u )2 + Á ( Dlu )2 + Á ( Da u )2
Ë ∂q ¯ Ë ∂L ˜¯ Ë ∂h ˜¯ Ë ∂g ˜¯ Ë ∂l ˜¯ Ë ∂a ˜¯
(9.41)
where the subscript “u” refers to the uncertainty in the parameter
and the partial derivatives are evaluated from Eq. 9.40. The uncer-
tainty in wavelength, Δλu is zero because the probe laser wavelength
does not change. Therefore, the accuracy in ΔT estimation depends
on the uncertainties in the estimates of η, α, g, and ΔΘ measurement.
∂T l Dq d
= (9.42)
dq 2p Ê aˆ
ÁË g - Lh
h ˜¯
∂T l qd DL
=- (9.43)
dL 2p Ê a ˆ L2
ÁË g - h
h ˜¯
∂T l qd 1
=- (9.44)
dh 2p Ê a ˆ h2
ÁË g - h ˜¯
∂T l qd 1
=- 2 (9.45)
da 2p Ê a ˆ Lh
2
ÁË g - h ˜¯
∂T l qd 1
= 2 Lh (9.46)
dg 2p Ê aˆ
ÁË g - h ˜¯
9.5 Michelson Interferometer

The experimental configuration of a Michelson interferometer
used for the measurement of laser cooling/heating of solids is
shown in Fig. 9.12. Fringes are produced by the superposition
of two coherent beams, of which one is a reference beam and the
other is a measurement beam. The latter either passes through or
gets reflected from the sample and its optical path length changes
whenever the sample is heated or cooled. So the fringe intensity,
measured by a detector, changes continuously with time because of
a change in the optical path length, ηL. Optical path length changes
because the sample length, L, and refractive index, η, are temperature
dependent. A He–Ne laser is separated into two beams by a beam
splitter, BS1, of which one is retroreflected by mirror M1 (reference
beam) and the other passes through the sample and is reflected
by mirror M2 (measurement beam). The path lengths from BS1 to
M1 and M2 are approximately equal. The retroreflected beams are
recombined by the same beam splitter and directed toward mirror
M3. After reflection from M3 the beams pass through a color filter,
CF, and a lens, which projects interference fringes on a screen. A
small wedge is created between the reflected wavefronts to produce
linear fringes [2]. A Pr3+-doped YAG sample is used for laser heating
studies. An argon ion laser beam resonantly excites the Pr3+ ion to its
excited levels [46]. Excited ions relax radiatively and nonradiatively
(Fig. 9.13) and the latter contributes to sample heating. The argon
ion laser beam is reflected by beam splitter BS2, which enters the
Michelson Interferometer 295
sample collinearly with the He–Ne laser. The color filter blocks
the heating beam from reaching the screen. Both the length and
the refractive index of a material change whenever it is heated. A
measurement of optical path length variation of the beam that
passes through the material reveals the sample temperature change.
So, prior knowledge of the temperature dependence of the material’s
refractive index and thermal expansion coefficient is needed [38].
The results obtained for Pr3+-doped YAG are summarized here [2].
He–Ne
Ar+
SA M3
BS2 BS1
M2
CF
Lens
SA sample Heat
BS beam splitter
M mirror Cool
D detector
CF color filter M1
Computer
Figure 9.12 Michelson interferometer setup used to measure laser-induced

heating of YAG:Pr3+.
3P
2
3
P0
1D
2
1
G4
617 nm
488 nm
3H
6
3H
4
Figure 9.13 Partial energy level diagram of Pr3+-doped YAG depicting

radiative and nonradiative relaxations.
9.5.1 Theory of Michelson Interferometer

The OPD of the two beams is given by
OPD = D1 – D2 = 2 η1(L1 – L2) +2η2L2 – 2η1L1
= 2L2(η2 – η1) = 2L2(η2 – 1) = 2L(η – 1) (9.47)
where η1 was set to 1.0 for air. Hereafter the sample length L2 and
index η2 will be replaced by L and η, respectively. L1 is the distance
from BS1 to the mirror (M1 or M2) and L2 is the sample length. The
temperature dependence of the refractive index and sample length
are given by Eqs. 9.8 and 9.9. L0 and η0 are the room-temperature
values of sample length and refractive index, respectively; hereafter
they are represented by L and η without the subscripts. The variation
in OPD is given by
Δ(OPD) =2[Δ(ηL) – ΔL] = 2[ΔηL + ηΔL – ΔL] (9.48)
After applying the procedure, described by Eqs. 9.8 to 9.12 the
above equation reduces to
Ê 1 ∂h aˆ
D (OPD) = 2Lh Á + a - ˜ DT (9.49)
Ë h ∂T h¯
where α and γ have their usual meanings, as explained before. If N
fringes translated for a given temperature change, ΔT, then Δ(OPD)
= λN. Equation 9.49 reduces to
Ê aˆ
l N = 2Lh Á g - ˜ DT (9.50)
Ë h¯
The temperature variation of the sample along the beam path is
given by
lN
DT = (9.51)
Ê aˆ
2Lh Á g - ˜
Ë h¯
From the above equation it is clear that ΔT is a function of N, L,
η, α, and γ. So the accuracy of ΔT estimate depends on the accuracy
with which these parameters were measured.
The YAG:Pr3+ sample length L0 is 0.01048 m, and its room-
temperature refractive index η0 is 1.823 at 632.8 nm [37], its linear
thermal expansion coefficient α is 7.8 × 10–6/K, the fractional change
in the optical path length γ is 11 × 10–6/K [38, 39], and the probe
laser wavelength is 632.8 nm. When 115 mW of a focused Ar+ beam

entered the sample the detector output was oscillatory, as expected,
and it approached the steady-state value in 500 s (Fig. 9.14). The
fringe period is fast (40 s) as soon as the Ar+ laser was turned on and
eventually it increased to 240 s for the last fringe. This variation in
the time period or chirping is due to the nature of sample heating.
As soon as the Ar+ laser is turned on, nonradiative relaxation sets
in, which heats the sample fast. Eventually the sample temperature
reached a steady-state value in about 500 s and the detector recorded
a total of 4.75±0.25 fringes (N) by this time. The number of waves
or fringes was measured manually from the plot. After substituting
these numbers in Eq. 9.51 the change in sample temperature along
the beam path is estimated to be 11.7±1.0 K [2]. The temperature
change is small because a small fraction of the incident power is
absorbed by the sample, which is a function of sample absorption
coefficient at the laser wavelength. The quantum efficiency of the
3P level is less than 1; in other words some of the excited ions relax
0
nonradiatively (represented by the wavy line in Fig. 9.13) to the
1D level, which contributes to heating. Moreover, the ions in the
2
lower levels 3H5, 3H6, 3F2, and 3F3 levels also undergo nonradiative
relaxation, which is represented by wavy lines. All these nonradiative
relaxation pathways contribute to a temperature change of 11.7±1.0
K. The uncertainty is large because of the following reasons. Using
a micrometer sample length can be measured to an accuracy of
10–5 m. The fringe intensity is measured by a detector whose
output is sinusoidal. From the signal plot the number of waves (or
fringes) can be measured to an accuracy of 0.25. The uncertainties
in the refractive index, the coefficient of thermal expansion, and
the fractional change in optical path length [38, 39] are 10–3, 0.4 ×
10–6/K, and 0.4 × 10–6/K, respectively. The uncertainty in the ΔT
calculation is similar that given by Eqs. 9.20 to 9.25.
The sample cools down if the argon ion laser beam is blocked.
The direction of fringe motion is expected to be opposite to that of
heating. This has been demonstrated experimentally and a sample
plot is depicted in Fig. 9.15. At time t = 0 s the Ar+ laser was on and
the detector output was varying, as expected. At time t = 135 s the
detector output was rising, and the Ar+ laser beam was blocked.
As expected the detector output started decreasing because the
material is cooling and the fringes moved in opposite directions.

During the cooling process the material slowly reached thermal
equilibrium and the detector output reached the steady-state value
in about 100 s after the Ar+ laser beam was blocked. The number of
fringes (or waves) measured depends on the Ar+ laser power and
whether the beam was focused in the material or not. To verify this
dependence an unfocused 488 nm laser beam and a He–Ne beam
were launched collinearly into the YAG:Pr3+ sample. As expected
the number of waves or fringes measured increased with the laser
power (not shown). Chirping is present in all the measurements.
This confirms that the plot shown in Fig. 9.14 is real and not due to
random fluctuations in the signal.
Figure 9.14 Temporal variation of the detector output obtained when 115
mW of an Ar+ laser beam was launched through the sample.
The sample was located at the focus. The Ar+ laser was on from
t = 0 to 600 s [2].
Figure 9.15 Temporal variation of the detector output during heating (Ar+
laser was on from t = 0 to 135 s) and cooling (Ar+ laser was off
from t = 135 to 300 s) periods [2].
The results of interferometric measurements are validated

by a thermocouple. For this purpose, the sample was mounted in
a metallic block, which was partly immersed in a water beaker
(Fig. 9.16). The glass beaker was heated on top of a hotplate. This
facilitates the sample heating and cooling in a controlled fashion. A
thermocouple was immersed into the water beaker at half height to
measure the water temperature. The sample holder assembly was
inserted along the path of the measurement beam of the Michelson
interferometer. The sample does not absorb the 632.8 nm beam and
hence there is no nonradiative relaxation in the sample. The sample
temperature was increased by heating the water. As expected the
detector output produced an interference pattern and produced five
fringes in 800 s (Fig. 9.17a). Thermocouple measurements revealed
a 14°C change in sample temperature (increased from 22.5°C to
36.5°C) during this time interval. The detector output is noisy
because the optical table was not isolated from floor vibrations and
the laser power was not stabilized. The change in sample temperature
is estimated using Eq. 9.51 as 12.3 ± 1.0°C, which is in reasonable
agreement with that measured by the thermocouple. When the
electric power was turned off to the hotplate sample temperature
was slowly decreasing, and the detector output revealed three
fringes in 2400 s (Fig. 9.17b). This corresponds to a temperature
change of 7.4 ± 0.8°C, which is in agreement with that measured
by the thermocouple, whose temperature decreased from 35°C to
27.6°C. The agreement was reproducible. During the cooling process
the water bath was allowed to cool to room temperature naturally
in ambient air, which was a slow process because the heat capacity
of water is a large value. Hence, there is perfect agreement between
the value given by the thermocouple and that estimated by Eq. 9.51.
This experiment confirms that the interferometric measurement of
temperature is reliable. The detector output is free from chirping
if the heating/cooling rate is constant. However, the agreement is
not perfect between the optical and thermocouple measurements
during sample heating, because of the procedure used in performing
the experiment and human error. The hotplate dial was set at 200°C
to heat the water. The temperature of the hotplate increased very
fast to reach the set value. However, there was some time lag (couple
of minutes) before the thermocouple senses higher temperatures,
because of the delay involved in heating the water. When the
temperature started to increase the computer was manually
triggered for data acquisition. This process has generated some
error in timing the events.
Figure 9.16 Sample mounting configuration for heating and cooling

studies. 1, YAG sample; 2, aluminum block; 3, glass beaker; 4,
water; 5 hotplate; 6, thermocouple.
We have demonstrated the use of the Michelson interferometer to

measure the sample temperature change due to laser beam heating.
An analytical model was developed to estimate the temperature
change that depends on fringe (or wave) counting. The passage
of a 115 mW focused laser beam through the sample raised its
temperature by 11.7 ± 1.0 K. Optical interferometric measurement
of temperature agreed with that measured by a thermocouple. This
method applies equally well to estimate the temperature changes in
laser cooling of solids also.
References 301
Figure 9.17 Temporal variation of the detector output measured when the
sample holder was (a) heated in a water bath from 22.5°C (t =
0 s) to 36.5°C (t = 853 s) and (b) naturally cooled in ambient air
from 35°C (t = 0 s) to 27.6°C (t = 2400 s).
References
1. Baker, B. (1998). Temperature Sensing Technologies (Microchip

Technology), AN679.
2. Farley, C. W., Reddy, B. R. (2011). Interferometric measurement of laser
heating in Praseodymium doped YAG crystal, Appl. Opt., 50, pp. 526–
531.
E. (1995). Observation of laser induced fluorescent cooling of a solid,
Nature, 377, pp. 500–503.
4. Soto, M. A., Nannipierri, T., Signorini, A., Lazzeri, A., Baronti, F., Roncella,
R., Bolognini, G., Pasquale, F. D. (2011). Raman-based distributed
temperature sensor with 1 m spatial resolution over 26 km SMF using
low repetition-rate cyclic pulse coding, Opt. Lett., 36, pp. 2557–2559.
5. Lapp, M., Hartley, D. L. (1976). Raman scattering studies of combustion,
Combust. Sci. Technol., 13, pp. 199–210.
6. Wade, S. A., Collins, S. F., Baxter, D. W. (2003). Fluorescence intensity
ratio technique for optical fiber point temperature sensing, J. Appl.
Phys., 94, pp. 4743–4756.
7. Chubb, D. L., Wolford, D. S. (2000). Rare-earth optical temperature
sensor, Rev. Sci. Instrum., 71, pp. 2233–2237.
8. Zhan, Y., Wu, H., Yang, Q., Xiang, S., He, H. (2008). Fiber Bragg grating
sensors for high temperature measurement, Opt. Lasers Eng., 46, pp.
349–354.
9. de Lima Filho, E. S., Nemova, G., Loranger, S., Kashyap, R. (2014).
Direct measurement of laser cooling of Yb:YAG crystal at atmospheric
pressure using a fiber Bragg grating, Proc. SPIE, 9000, p. 90000I.
10. Hariharan, P. (2003). Optical Inteferometry (Academic Press, San
Diego, USA).
11. Yoshizawa, T. (2009). Handbook of Optical Metrology: Principles and
Applications (CRC Press, Boca Raton, USA).
12. Wilkie, D., Fisher, S. A. (1963). Measurement of temperature by Mach-
Zehnder interferometry, Proc. Inst. Mech. Eng., 178, pp. 461–470.
13. Sandhu, S. S., Weinberg, F. J. (1972). A laser interferometer for
combustion, aerodynamics and heat transfer studies, J. Phys. E: Sci.
Instrum., 5, pp. 1018– 1020.
14. Steel, W. H. (1983). Interferometry (Cambridge University Press, New
York, USA).
15. Malacara, D. (1992). Optical Shop Testing (John Wiley & sons, New
York, USA).
16. Toal, V. (2012). Introduction to Holography (CRC Press, New York,
USA).
17. Hariharan, P. (1969). Modified Mach-Zehnder interferometer, Appl.
Opt., 8, pp. 1925–1926.
18. Kowalski, F. V., Teets, R. E., Demtröder, W., Schawlow, A. L. (1978). An
improved wavemeter for cw lasers, J. Opt. Soc. Am., 68, pp. 1611–1613.
19. Lee, L. S., Schawlow, A. L. (1981). Multiple-wedge wavemeter for
pulsed lasers, Opt. Lett., 6, pp. 610–612.
References 303
20. Luo, H., Sun, Q., Xu, Z., Liu, D., Zhang, L. (2014). Simultaneous
measurement of refractive index and temperature using multimode
microfiber-based dual Mach–Zehnder interferometer, Opt. Lett., 39,
pp. 4049–4053.
21. Lu, P., Chen, Q. (2010). Asymmetrical fiber Mach–Zehnder
interferometer for simultaneous measurement of axial strain and
temperature, IEEE Photonics J., 2, pp. 942–952.
22. Dullo, F. T., Sokolov, V., Chauvet, C., Lindecrantz, S., Solbø, S. A., Hellesø,
O. G. (2015). Application of Mach–Zehnder interferometer based
on long period grating structure for salinity measurement in water
environment, Proc. SPIE, 8988 89881T-1.
23. Ashrafi, Z. N., Ashjaee, M., Askar, M. H. (2015). Two-dimensional
temperature field measurement of a premixed methane/air flame
using Mach–Zehnder interferometry, Opt. Commun., 345, pp. 55–63.
24. Koechner, W., Bass, M. (2003). Solid-State Lasers (Springer, New York,
USA).
25. Moerner, W. E. (1988). Persistent Spectral Hole Burning: Science and
Applications, Topics in Current Physics, Vol. 44 (Springer, New York,
USA).
26. Epstein, R. I., Sheik-Bahae, M. (2009). Optical Refrigeration: Science and
Applications of Laser Cooling Of Solids (Wiley VCH Verlag, Weinheim).
27. Dieke, G. H. (1968). Spectra and Energy Levels of Rare-Earth Ions (John
Wiley & Sons, USA).
28. Weber, M. J. (1967). Probabilities for radiative and nonradiative decay
of Er3+ in LaF3, Phys. Rev., 157, pp. 262– 272.
29. Riseberg, L. A., Moos, H. W. (1967). Multiphonon orbit-lattice relaxation
in LaBr3, LaCl3 and LaF3, Phys. Rev. Lett., 19, pp. 1423–1426.
30. Kamma, I., Kommidi, P., Reddy, B. R. (2008). Design of a high
temperature sensing system using luminescent lifetime measurement,
Rev. Sci. Instrum., 79, p. 096104.
31. Reddy, B. R., Kamma, I., Kommidi, P. (2013). Optical sensing techniques
for temperature measurement, Appl. Opt., 52, pp. B33–B39.
32. Messias, D. N., Vermelho, M. V. D., Gouveia-Neto, A. S., Aitchison, J.
S. (2002). All optical integrated upconversion fluorescence based
point temperature sensing system using Er3+ doped silica-on-silicon
waveguides, Rev. Sci. Instrum., 73, pp. 476–479.
33. Paez, G., Strojnik, M. (2003). Erbium-doped optical fiber fluorescence
temperature sensor with enhanced sensitivity, a high signal-to-noise
ratio and a power ratio in the 520-530- and 550-560 nm bands, Appl.
Opt., 42, pp. 3251–3258.
34. Paez, G., Strojnik, M. (2003). Experimental results of ratio-based
erbium-doped-silica temperature sensor, Opt. Eng., 42, pp. 1805–
1811.
35. Patterson, W., Soto, E., Fleharty, M., Sheik-Bahae, M. (2007).
Differential luminescence thermometry in laser cooling of solids, Proc.
SPIE., 6461, p. 64610B.
36. Farley, C. W., Reddy, B. R. (2011). Mach-Zehnder interferometric
measurement of laser heating/cooling in Yb3+: YAG, Proc. SPIE, 7951,
p. 795101.
37. YAG data sheet, VLOC, FL (www.vloc.com).
38. Fan, T. Y., Daneu, J. L. (1998). Thermal coefficients of the optical path
length and refractive index in YAG, Appl. Opt., 37, pp. 1635–1637.
39. Wynne, R., Daneu, J. L., Fan, T. Y. (1999). Thermal coefficients of the
expansion and refractive index in YAG, Appl. Opt., 38, pp. 3282–3284.
40. Sommergen, G. E. (1981). Optical heterodyne profilometry, Appl. Opt.,
pp. 610–618.
41. Huang, C. C. (1984). Optical heterodyne profilometer, Opt. Eng., 23, pp.
365–370.
42. George, M. C., Reddy, B. R., Jagannath, H., Perera, G. M., Venkateswarlu,
P. (1991). Measurement of surface roughness using a noncontact
profilometer, in Laser Spectroscopy and Nonlinear Optics of Solids,
Radhakrishna, S., Tan, B. C., eds. (Springer Verlag/Narosa, New Delhi,
India), pp. 147–152.
43. Kim, Y. K., Reddy, B. R., George, T. G., Lal, R. B. (1998). Optical
heterodyne interferometry technique for solution crystal growth rate
measurement, Opt. Eng., 37, pp. 616–621.
44. Reddy, B. R. (2014). Applications of optical interferometer techniques
for precision measurements of changes in temperature, growth and
refractive index of materials, Technologies, 2, pp. 54–75.
45. Taylor, J. R. (1997). An Introduction to Error Analysis: The Study of
Uncertainties in Physical Measurements, 2nd Ed. (University Science
Books, Sausalito, CA).
46. Kamma, I., Kommidi, P., Reddy, B. R. (2009). High temperature
measurement using luminescence of Pr3+ doped YAG and Ho3+ doped
CaF2, Phys. Status Solidi C, 6, pp. S187–S190.
Chapter 10
Fluoride Glasses and Fibers
Mohammed Saad
Thorlabs, Inc., 56 Sparta Avenue, Newton, NJ 07860, USA
msaad@thorlabs.com
10.1 Introduction
During the last three decades, optical fibers have experienced
an extraordinary and rapid development. They are being used, as
passive and active mediums, in various industrial applications,
in fields as diverse as telecommunication, medicine, aerospace,
defense, spectroscopy, sensing, laser power delivery, fiber lasers, etc.
Optical fibers are now critical parts in many high-tech devices.
Of course fiber parameters have to be tailored to each application
needs. Unfortunately, there is no single fiber material that can fulfill
all types of application needs. Engineers and researchers have to
make some compromises when choosing the right fiber material
for the right application. There are some important criteria that
one has to consider when choosing an optical fiber to be used in a
specific application. The first criterion is technical. The glass should
have the required properties, such as transmission window, glass

www.panstanford.com
306 Fluoride Glasses and Fibers
characteristic temperatures, and durability. The second criterion

is economical. The technology should be mature enough to provide
high-quality optical fibers at an effective cost with a reasonable
yield.
Silica fiber technology is the most establish technology so
far. Billions of dollars have been spent to develop this technology
and bring it to the current level. This was mainly motivated by the
huge telecommunication market. Standard silica optical fibers,
used in the telecommunication field, are produced in thousands of
kilometers, with high mechanical strength and ultralow loss. The
fiber loss is approaching the theoretical value, which means that
the material has ultrahigh purity and the production process is well
under control. However, these fibers cannot be used for application
above 2 microns since the material is opaque. More recently, silica
photonic crystal fibers (PCFs), with some special design, negative
curvature fiber [1], have been shown to transmit at few wavelengths
over 2 microns. But these fibers are still at the R&D phase. Of course,
applying the same design to materials that are already transmitting
in the midinfrared region will enlarge the transmission spectrum to
longer wavelengths for infrared materials, too.
For applications requesting fibers with either transmission
above 2 microns or continuous transmission from the ultraviolet
to the midinfrared region, there are few infrared materials that can
be considered, such as heavy metal fluorides [2], chalcogenides [3],
heavy oxides such as tellurite [4], phosphates [5], single crystals
such as sapphire [6], and polycrystalline materials. These fibers are
commonly called exotic fibers. Again, one has to choose among these
materials according to the application requirements. However, few
of these materials can be drawn into high-quality optical fibers that
can meet industrial application needs.
Amorphous materials have definitely some advantages when
compared to crystalline materials. Fibers made from amorphous
materials can have a core cladding structure and have low loss. They
are also more robust and very flexible. They can also be drawn into
long-length fibers. Furthermore, fiber parameters can be tailored to
application needs by adjusting core and cladding compositions.
Among all exotic materials, fluoride glass fiber technology is the
more advanced technology so far [7]. In fact, in the last two decades,
significant progress has been achieved and the technology is now
Bulk Glass 307
mature enough to meet many industrial application requirements.

Fluoride fibers are now being used in some industrial applications,
including the most demanding ones, such as defense and aerospace.
Almost all different types of fibers have been produced with
fluoride glasses, including doped and undoped multimode and
single-mode fibers. Some exotic shapes such as hexagonal, square,
and D-shaped fibers have been reported as well.
Beside the state of the technology, fluoride glasses show also
outstanding optical properties that make them the material of
choice for many infrared and even multispectral applications. They
have a low refractive index and a low and negative dn/dt. They can
be heavily doped and co-doped with any rare earth ions for active
applications, such as fiber lasers and amplifiers. Stable fluoride
glass composition can contain up to 100,000 ppm of rare earth ions.
This concentration is ranging from only a few hundreds to a few
thousands of ppm in silica fibers. In fact, lanthanum fluoride is one of
the elements entering some fluoride glass compositions, and it can
be substituted by any other rare earth ion without compromising
glass quality. Consequently, compact fiber laser can be made using a
short length, a few centimeters, of doped fluoride glass fiber.
10.2 Bulk Glass

Glass formation rules established for oxide glasses cannot be used to
predict glass formation in exotic glass systems and especially ionic
glasses such as halide glasses. So, nobody was expecting that glass
formation can be that rich in complex halide systems.
The first fluoride glass has been reported based on beryllium
fluoride (BeF2) glass systems, as reported by Van Uitert [8]. This glass
has a low refractive index (1.30) and a glass transition temperature
of 250°C. However, these glasses were never used for any practical
applications, mainly because their infrared transmission is not that
far from silica and also because beryllium fluoride is very toxic.
In the midseventies, a new family of fluoride glasses was
accidentally discovered at Rennes University by Michel Poulain
[9]. Michel was working on crystalline phases of BaZrf6 and LnZrF7
systems. He was mixing different rare earth fluorides with zirconium
fluoride (ZrF4) and barium fluoride. It turned out that the structure
of some crystals showed quite big free sites. After calculation, it

turned out that the free sites can contain an anion such as sodium
(Na). So Poulain decided to add sodium fluoride to the mixture, to see
if the sodium ions can occupy the free sites. He mixed the chemicals
in a celled nickel tube under vacuum and placed the celled tube in a
furnace at 800°C. The day after he took the nickel tube and cooled it
rapidly by doping it into cooled water at room temperature. So the
melt was quenched from 800°C to room temperature very quickly.
Poulain was surprised to see that the mixture was not anymore a
powder, as he was expecting, but a piece of glass. The first heavy
metal fluoride glass was born! He obtained in fact a colored glass.
After analysis, it turned out that the sample is amorphous and has
a wider transmission than silica glasses. The real fluoride glass
adventure started!
Since then, tens of research laboratories across the world
have participated in the development of fluoride glass and fiber
technology. Hundreds of glass systems have been reported. And
scientists realized that glass formation in halide systems is more
common than expected. More interestingly, the glass-forming area is
quite large, so some glass parameters can be tailored for each specific
application need. This includes refractive index n, glass transition
temperature Tg, the coefficient of thermal expansion (CTE), etc.
Three main glass families have been considered representative
of fluoride glasses in general: fluorozirconate, fluoroaluminate, and
fluoroindate. They are, respectively, based on zirconium fluoride
(ZrF4), aluminum fluoride (AlF3), and indium fluoride (InF3).
Generally, the glass family name will use the name of the chemical
that has the higher molar percentage in the composition. Therefore,
fluorozirconate is used for zirconium fluoride (ZrF4) base glasses,
fluoroaluminate for aluminum fluoride (AlF3) base glasses, and
fluoroindate for indium fluoride (InF3) base glasses. Indium fluoride
glasses have the largest transmission window, and aluminum
fluoride glasses have the highest resistance to liquid water. In fact,
some aluminum fluoride glasses can be placed in hot water for many
days without showing any degradation.
ZBLAN and ZBLA are two standard compositions reported in
the fluorozirconate glass family. These two compositions have been
widely used for almost all research programs and applications
involving fluoride glasses. In fact, they are the most stable
Bulk Glass 309
compositions in all fluoride glass history. This doesn’t mean that

they are the only composition that can be drawn into high-quality
optical fibers. We have been able to draw into fibers many different
compositions in different glass systems.
Like any multicompound glass, the preparation process is very
critical. In fact, using the same composition one can end up with bad
or good glass, depending on the raw materials’ quality and which
preparation process was used. There are as many preparation
processes as research teams that were involved in fluoride glass
research. Therefore, tremendous work has been carried out to
optimize the fluoride glass preparation process [10–13]. Of course,
and it’s not surprising, that the preparation process has to be
optimized for each glass family and even in some cases for each glass
composition.
More complex systems, using a mix of halide ions such as
chlorides and fluorides [14–17] have been also explored and quite
stable compositions were reported. However, these glasses are not
stable enough to be drawn into high-quality optical fibers, or can’t
be used in any practical application, because they are hydroscopic.
Nevertheless, the study of these systems can provide us with
interesting data that help scientists understand glass formation in
exotic systems.
Stable compositions, with some interesting properties, have been
also reported in aluminum fluoride [18] and indium fluoride glass
systems [19]. These compositions contain quite high concentrations
of thorium fluoride. Because of the radioactivity of thorium, these
glasses have never been used in any application. Some of these glasses
have very interesting and unique properties, such as ultraresistance
to water attack. The glass can be left in hot water for days without
showing any degradation. Aluminum fluoride glasses containing
20 molar percent of thorium fluoride (ThF4) have a low refractive
index, too, around 1.45; a high glass transition temperature around
420°C; and a low CTE, around 140 × 10–7 K–1. Thus, only few research
groups have investigated these systems. The stable composition in
this system is BATY, containing 20 molar% BaF2, 28.7% AlF3, 22.6 %
ThF4, and 28.7% YF3.
In general, quite stable compositions contain minimum three
compounds can be cast into bulk samples with a few millimeters
thickness. Most of these compositions cannot be drawn into optical
fibers. Glass stability is most often drastically improved by adding

few percent of one or two additional fluorides. As an example, the
addition of 3% to 4% AlF3 to two standard zirconium fluoride glasses
ZBN (ZrF4, BaF2, NaF) and ZBL (ZrF4, BaF2, LaF3) has improved
considerably the glass stability [20]. Glass samples with a few
centimeters thickness can be obtained. Therefore, the most stable
compositions in fluorozirconate glass systems contain 3% to 4%
of aluminum fluoride. Table 10.1 shows the composition in molar
percent of the two more stable compositions in the fluorozirconate
family. These two compositions have been widely used in all studies
and applications concerning fluoride glasses.
Table 10.1 Standard fluorozirconate glass, ZBLAN and ZBLA, compositions

in molar%
ZrF4 BaF2 LaF3 AlF3 NaF

ZBLAN 53 20 4 3 20
ZBLA 57 34 5 4
10.2.1 Glass Preparation

The modified chemical vapor deposition process (MCVD) has shown
to be the process of choice to produce high-quality optical fibers
in silica fiber technology. This process can be used when starting
materials entering glass composition have quite similar physical and
chemical properties. The MCVD process allows us to use ultrahigh-
purity materials and also more importantly to keep the materials
clean during the process. Thus, the impurities’ contribution to the
loss has become negligible. Consequently, the commercial-standard
silica fibers have attenuation close to the theoretical value.
Unfortunately, the MCVD process cannot be used to prepare
most of the multicompound glasses, including fluoride glasses. In
fact, stable fluoride glass compositions contain minimum four to five
chemicals, with physical and chemical properties quite different.
Thus, it is difficult and even impossible to process all starting
materials, entering the glass composition, at the same range of
temperature and pressure. However, binary thin films have been
reported by Jacoboni et al. [21] using chemical vapor deposition.
Bulk Glass 311
This process is only limited to binary thin films and can’t be used to
prepare bulk glasses and preforms.
As far as fluoride glasses are concerned, the preparation process
commonly used is melting and casting. In this case ultrahigh-purity
starting materials are required. And raw materials’ purity is a big
concern.
In fact, many chemical molecules and ions, such as transition
metals [22], rare earth elements [23], water [24], CO2 [25], and
some organic molecules, have their fundamental absorption in the
transmission window of fluoride glasses. And only a very small
amount of these impurities in the glass will have a huge absorption
peak in the fiber spectrum. Thus, from the beginning the purity of
raw materials was a critical issue. At that time no ultrahigh-purity
raw materials were available, especially those entering stable
glass compositions. As an example, Table 10.2 shows attenuation
induced by some transition metals and rare earth elements [22]. As
an example, –OH and transition metal ion concentration has to be
kept in the ppb level. There are two challenges here: The first one
is to be able to produce the material with a low impurity level, and
the second is to be able to measure this impurity level with good
accuracy. Another challenge is that many raw material suppliers are
not really interested in making some chemicals with such ultrahigh-
purity materials since the market is too small. So sometimes you
have to produce or purify your own raw materials.
Table 10.2 Attenuation (dB/km/ppm) induced by some transition metals

and rare earth elements
Elements 2 mm 2.5 mm 3 mm
Transition metals (dB/km/ppm) (dB/km/ppm) (dB/km/ppm)
Fe 90 28 2
Co 130 31 4
Ni 90 6 0.5
Cu 3 0.14 0.01
Rare earth
Ce – – 7.7
Pr 27 1.8 0.4
Tb 25 – 14.3
Various research programs have been dedicated to develop

ultrasensitive analysis methods and preparation processes to
produce and characterize ultrahigh-purity starting materials,
especially those entering stable glass compositions. Since the most
stable fluoride glass compositions have been reported in zirconium
fluoride glasses, these programs have focused first on ZrF4, HfF4,
NaF, BaF2, AlF3, and rare earth elements LnF3 [26, 27].
Three preparation processes have been developed to prepare
fluoride glasses; only the two first ones have been widely used [28]:
∑ Direct melting of anhydrous fluoride, in platinum, gold, or
glassy carbon crucibles, under a dry atmosphere. The melting
temperature can vary from 800°C to 1000°C and the meting
duration from one to three hours. At the end of the melting
step, the glass is cooled down to around 650°C to 700°C and
then poured into a metallic mold preheated at a temperature
lower but close to the glass transition temperature. The mold
is then cooled down to room temperature in a few hours. Once
the mold is at room temperature, the glass is de-molded and
polished using a standard polishing process.
∑ The ammonium bifluoride process, which consists of using
fluoride and oxides as starting materials. The oxides are
converted into fluoride by ammonium bifluoride (NH4FHF)
during the fluorination step prior to melting [29]. The
fluorination process takes place at a temperature ranging
from 300°C to 400°C during two to three hours, depending
on the composition and glass family. Once the fluorination is
completed, he glass is obtained by melting the mixture under a
dry and controlled atmosphere. The melting temperature and
schedule are similar to the process above. The fluorination
reactions of zirconium oxide and indium oxide are given
below as an example.
ZrO2 + 2NH4FHF ¤ ZrF4 + 2H2O + 2NH3
In2O3 + 3NH4FHF ¤ 2InF3 + 3H2O + 3NH3
Ammonium bifluoride’s reaction with oxides produces water.
Thus, one has to be sure that water is removed during the
glass preparation process. In fact water has its fundamental
absorption at 2.87 microns in a fluoride glass matrix. This is
Glass Characterization 313
too close to the minimum attenuation wavelength, 2.6 microns.

G. Fonteneau has reported that 1 ppm of water induces 5000
dB/km [30]. Different reactive atmospheres (RAP) have been
also used, during glass processing, to remove water traces in
the final glass. HCl, CCl4, and NF3 [31] have been widely used.
∑ The sol gel process has been also reported by Mohammed
Saad and Marcel Poulain [32]. This process consists of
using organometallic chemicals as starting materials. These
chemicals are mostly liquid at room temperature, and thus,
they can be easily purified compared to solid raw chemicals.
In fact, a liquid can be purified using and combining different
purification methods that can’t be used for solid chemicals,
such as distillation, phase separation, and extraction.
First we mix chemicals of ions entering the glass composition.
After hydrolysis we obtain a transparent oxide gel. The gel is then
dried to eliminate the excess of organic solvent and water. The dried
gel is fluorinated using gaseous HF. The fluoride glass is obtained by
melting the fluorinated gel in a platinum crucible and casting it into
a brass mold.
10.3 Glass Characterization

Once the glass samples are cast and de-molded, they go through
different characterization processes. The glass transition
temperature Tg, glass crystallization temperature Tx, and melting
temperature Tm are obtained using differential scanning calorimetry
(DSC). The thermal expansion coefficient α is obtained by
thermomechanical analysis (TMA). Glass transmission spectra are
obtained using ultraviolet, visible, and infrared spectrometers. And
the refractive index is obtained by the Abbe refractometer as well as
other methods using different laser wavelengths.
10.3.1 Differential Scanning Calorimetry

The DSC technique is used to determine glass’s characteristic
temperatures. It is also used to determine some kinetic and
crystallization parameters, such as the glass critical rate. Figure
10.1 shows a typical DSC curve of a glass sample. This curve is
obtained by heating, at a given heating rate, 35 to 40 mg of glass

and a reference in a microfurnace. Exothermic and endothermic
heat flows are plotted as a function of the temperature or time. So,
we can determine the glass transmission temperature Tg, the glass
crystallization temperature Tx, and the glass melting temperature
Tm.
Figure 10.1 Typical DSC scan of a glass sample.
DSC is also used to estimate the stability of glass and crystalliza-

tion parameters. Some stability criteria have been proposed to esti-
mate glass stability. These criteria use characteristic temperatures:
glass transition Tg, glass crystallization Tx, crystallization peak Tp,
and glass melting Tm. Ruby was the first one who proposed a crite-
rion to estimate glass stability using DSC measurement data [33].
This criterion is described by the equation below:
Hr = (Tx – Tg)/(Tm – Tx)
A stable glass will have a small (Tm – Tx) and a large (Tx – Tg).
That means a crystallization peak close to melting temperature and
far from glass transition temperature.
The Hr criterion has been simplified to Hr¢; in fact most of the
time the melting temperature is not well defined in multicompound
systems.
Hr¢ = (Tx – Tg)/Tg
Saad and Poulain proposed a new stability criterion S in 1985
[34]. The new criterion is defined as below:
S = (Tx – Tg)(Tp – Tx)/Tg
The new criterion takes into account the crystallization peak

shape. A stable glass will have its crystallization temperature far
from its glass transition temperature (large (Tx – Tg)), and a wide
crystallization peak (large (Tp – Tx)). The S criterion is more sensitive
to small composition changes than Hr is. Consequently, it is more
useful for composition optimizations. In fact, a small composition
change will not affect significantly the characteristic temperatures
but can significantly affect the stability of the glass. Figure 10.2
shows a comparison of the three stability criteria with a composition
change in 69 ZrF4; (23 – x) BaF2; (6 + x) SrF2; 2AlF3.
Figure 10.2 Comparison of H¢, Hr, and S glass stability criteria.
Of course the ultimate way to determine glass stability is to

determine the thermodynamic and kinetic parameters of each
composition. We can also determine the critical cooling rate (CCR);
this is the slowest cooling rate at which one can cool down a glass
from the liquid state without crystallization.
The CCR can be calculated from time-temperature-transformation
(TTT) curves [35]. This method is time consuming and quite
complicated to use. Indeed, many measurements are required to
plot TTT curves. Furthermore, TTT curves can only be obtained
for quite stable compositions. Generally, glass stability criteria are
used when exploring the glass formation region in a new system.
The DSC duration is about 20 to 25 minutes. Furthermore, we can
use disposable aluminum pans to perform the measurement, rather
than small platinum crucibles, which are difficult to clean. When
the glass formation region and the stable composition region have
been defined using the stability criterion, one can use the CCR to
determine the most stable composition of the system among all stable
compositions. Table 10.3 shows the CCR values for the two standard
fluorozirconate glasses ZBLAN and ZBLA. The ZBLAN composition
has the lowest CCR among all fluoride glass compositions reported
so far [36]. In fact, after melting, the ZBALN composition glass can be
cooled slowly in the crucible and no crystallization occurs.
Table 10.3 Critical cooling rates for ZBLAN and ZBLA
Glass Rc [°C/s]
ZBLAN 0.13
ZBLA 0.28
Table 10.4 shows some properties of some fluoride glass families.

The glass transition temperature varies between 250°C and 450°C,
the CTE varies between 140 and 200 × 10–7 K–1, and the refractive
index in the visible varies between 1.45 and 1.60. Consequently, no
antireflective coating is required for many applications. The Fresnel
reflections are low in the 3% to 4% range per end face in the visible
and even less in the infrared.
Table 10.4 Some fluoride glass properties
Glass Glass Transition Refractive index Coefficient of Thermal

Family Tg (°C) at 589.3 nm expansion 10–7K–1
ZrF4 230–310 1.48–153 150–200
InF3 290–320 1.48–1.53 170–190
AlF3 360–420 1.42–1.46 150–180
ZnF2 280–320 1.50–1.54 170–200
CdF2 300–360 1.48–1.50 140–160
10.3.2 Glass Transmission

Fluoride glasses have a large transmission window. They are
transparent from the ultraviolet to the midinfrared region without
any absorption peak. The transmission window depends on the
glass family and composition. The transmission is limited in the

ultraviolet by electronic transition and in the infrared by the
multiphonon. Figure 10.3 shows a comparison between zirconium
and indium fluoride glass transmissions.
Fluoride Glass Transmission (Including Fresnel Reflections)

100
90
80
Transmission (%)
70
60
50
40
30
20
10
0
0 1 2 3 4 5 6 7 8 9 10
Wavelength (mm)
Figure 10.3 ZrF4 and InF3 fluoride glass transmissions (9 mm thickness).
Indium fluoride glasses have a larger transmission window

than zirconium fluoride glasses. This is due to the phonon energy
of indium fluoride glasses, which is lower than that of zirconium
fluoride glasses.
Both glass families offer a large choice of compositions. So glass
parameters can be adjusted to meet different application needs.
Fluoride glasses offer many advantages to be used for bulk
optics, such as lenses, filters, and windows. We demonstrated that
aspheric lenses can be obtained by molding and diamond point
turning techniques. Figure 10.4 shows a photo of large lenses of
ZBLAN obtained by the diamond point turning technique. The
surface roughness is 2 nm.
10.3.3 Refractive Index and Dispersion

The refractive index is an important parameter for the optical
designer to carefully design optical systems and optical waveguides.
In fluoride glass systems, this parameter can be linearly adjusted
to each application need by adjusting the composition. In fact
most of the glass systems have a quite wide glass formation area.
Furthermore, large substitution of any ion by another from the
same column of the periodic table of the element is possible. As an

example, ZrF4 can be totally replaced by hafnium fluoride (HfF4). We
can go from the ZBLAN to the HfBLAN composition, where 53% of
ZrF4 is totally replaced by HfF4.
Figure 10.4 ZBLAN convex and concave lenses made by the single-point
diamond turning technique.
The refractive index of fluoride glasses is ranging from 1.45 to

1.60 at the sodium D wavelength 589 nm.
Lipeng Zhang et al. have studied relationships between the
refractive index and dispersion of fluoride glasses and their
compositions [37]. They found a good agreement between the
calculated and measured values. The equation used to calculate
the refractive index at different wavelengths is given below, where
li and lj are, respectively, the ultraviolet and infrared absorption
wavelengths of the glass, fi and fj are the constants related to
oscillator strength.
k m
nl1 - 1 = Â
i =1
fi l 2 (l 2 - li2 ) + Â f l (l
j =1
j
2 2
- l 2j ) (10.1)
One has to take into consideration glass processing, especially

when glass compositions contain chemicals that can sublimate
during glass preparation.
Fluoride glasses show low dispersion, lower than silica glasses.
The Abbe number is about 75. The Abbe number is given by the
equation below:
VD = (nD.1)/(nF – nc) (10.2)

where nD is the refractive index at the D sodium line 589.3 nm, nF is
the refractive index at the hydrogen blue line 486.1 nm, and nC is the
refractive index at the hydrogen red line 656.3 nm.
The refractive index at different wavelengths can be obtained
also using the Sellmeier empirical equation, where n is the refractive
index and mi are the Sellmeier coefficients, which are determined
experimentally.
m0 l 2 m1 l 2 m2 l 2
n( l ) = + + +A (10.3)
l 2 - m32 l 2 - m42 l 2 - m52
Table 10.5 and Table 10.6 show, respectively, the refractive

indexes of ZBLAN and indium fluoride glasses measured at different
wavelengths.
Table 10.5 ZBLAN glass measured refractive index at different wavelengths
ZrF4 Glass
Wavelength l (nm) Refractive index nl
405 1.506
486.1 1.504
587.6 1.4995
632.8 1.49807
656.3 1.4974
1060 1.49198
1550 1.48883
3500 1.47447
4500 1.46312
Table 10.6 InF3 glass measured refractive index at different wavelengths
InF3 Glass
Wavelength l Refractive index nl
4500 1.45886
2000 1.47384
1064 1.47828
587.6 1.48545
485 1.49005
404.7 1.49645
Using the Sellmeier equation one can calculate the refractive

index at any wavelength. Figure 10.5 shows the refractive index of
ZBLAN glass. The Sellmeier coefficients are given in Table 10.7.
Figure 10.5 Dispersion of ZBLAN glass.
Table 10.7 Sellmeier coefficients for ZBLAN glass
Coefficient Cladding
m0 0.705674
m1 0.515736
m2 2.204519
m3 0.087503
m4 0.087505
m5 23.80739
A 1
Figure 10.6 displays the refractive index of indium fluoride (InF3)

glass at different wavelengths. The Sellmeier coefficients are given in
Table 10.8.
Optical Fiber 321
Figure 10.6 Dispersion of indium fluoride glass.
Table 10.8 Sellmeier coefficients for indium fluoride glass
Coefficient Cladding
m0 0.68462594
m1 0.4952746
m2 1.4841315
m3 0.0680833
m4 0.11054856
m5 24.4391868
A 1
10.4 Optical Fiber

Fluoride glasses are, in fact, the only material that transmits light
from the ultraviolet up to the midinfrared and can be drawn into
high-quality optical fibers. Furthermore, as they are transparent in
the visible region, one can use visible light and standard equipment
for troubleshooting and optical alignment.
10.4.1 Fiber Drawing

Fluoride glass fibers are produced using the same drawing technique
as the one used for silica fibers. They are drawn from solid preforms.
This process is very effective and allows good control of fiber
parameters such as fiber diameter and core cladding concentricity.
Furthermore, fibers with exotic design can also be produced using
this process, such as hexagonal, D-shaped, and square fibers, and
PCFs [38].
There are few soft glass drawing towers in the world, but these
are all different. In fact, they are all homemade. Generally, they are
similar to the silica one but shorter. This is due to the fact that glass-
drawing temperatures for soft glasses are much lower than those of
silica. Consequently, the distance between the drawing furnace and
the coating cup can be kept much shorter than with silica.
The drawing furnace is the more important piece of equipment,
since it is the equipment especially designed for fluoride or soft
glass fiber drawing. All other equipment pieces, fiber diameter
measurement, coating cup, UV lamp, etc., are exactly the same as
those used on the silica drawing tower. However, some settings have
to be adjusted according to soft glass drawing parameters, which
are different from those used for silica. The drawing feed, drawing
temperature, drawing speed, and coating pressure ultraviolet lamp
power are all different. Fluoride glass preforms are also much
smaller than silica ones. The preform diameter can vary from 8 to 16
mm and 100 to 150 mm in length.
The double-crucible technique, which has been developed to
draw low–melting temperature oxide glasses, has been also used to
draw fluoride fibers [39]. A few hundreds of meters of single-mode
and multimode fibers have been obtained. This technique is effective
to draw long-length fibers but suffers from poor core cladding
concentricity and crystal and bubble formation at the core cladding
interface.
Figure 10.7 shows a typical drawing tower for soft glass. It is
constituted with a translation table with micron accuracy to feed the
preform into the drawing furnace, a drawing furnace, fiber diameter
measurement with submicron accuracy, a coating cup, concentricity
measurement, and an ultraviolet lamp. All these pieces of equipment
are controlled by the computer.
Optical Fiber 323
Figure 10.7 Soft glass drawing tower.
Almost all types of fibers have been produced with fluoride

glasses. Single-mode and multimode fibers are prepared with good
geometrical, optical, and mechanical properties. For example, 125
micron fibers are obtained with a diameter variation of ±1 micron.
Figure 10.8 shows fiber diameter variation during the drawing
process of a 125 micron indium fluoride fiber. The diameter is
automatically controlled.
Figure 10.8 A 125 um InF3 fiber’s diameter variation during the drawing
process.
Different methods are used to make fluoride glass preforms. The

easiest one is built-in casting [40]. This technique consists of pouring
clad glass into a preheated vertical mold. A few seconds later the
bottom of the brass mold is opened, and the unsolidified glass flows
out. We obtain a clad tube. The core is then poured into the cladding
tube. This technique is quite simple and yields fiber with quite good
attenuation and very low –OH content. The main advantage of this
technique is that the core cladding interface is never exposed to
the laboratory atmosphere. The only drawback is we end up with a
tapered core. This technique cannot be used to produce single-mode
fibers. In fact, it is very difficult to prepare a preform using the built-
in casting technique, with a core cladding ratio of 0.072, which is the
core cladding ratio for a 9 /125 single-mode fiber.
Another technique used to prepare fluoride glass preforms is
rotational casting, proposed by Tran [41]. First we produce a tube
with cladding glass by spinning the cladding glass at 3500 to 5000
rpm in a rotational jig. Then the core is poured into the tube. In
this case also the core cladding interface is never exposed to the
laboratory atmosphere. The core tapering issue is eliminated or
drastically reduced. The mechanical design of the jig is very critical.
One has to avoid any mold vibration during the spinning step. In fact,
any mold vibration will result in as off-center tube.
The third technique to prepare performs is the rod-in-tube
technique. This technique is similar to rotational casting, but the
core is not poured into the cladding tube. The core is a solid rod
that is inserted into the tube prior to drawing. In this case, the core
cladding interface is exposed to the laboratory environment; thus
care has to be taken to avoid scratches and contamination, especially
with water. The insertion step is critical, too. One has to avoid
scratching the inner surface during this step.
The rotational casting technique is also used to produce cladding
tubes with a small inner diameter, required to produce single-mode
fibers.
For some specific fibers, one has to combine some or all these
techniques to be able to produce some specific fibers.
Figure 10.9 shows a 9 micron core single-mode fiber and a 450
micron core multimode fiber. The multimode fiber is made using the
built-in casting method, and the single-mode fiber is made using the
rod-in-tube process. Multiple drawings are necessary to draw the
Optical Fiber 325
final single-mode fiber. The number of drawings depends on the

fiber core dimensions. The two types of fibers have been cleaved
using a standard cleaver with adjusted parameters. The two cleaves
have been obtained using a Vytran cleaver. They look very good,
meaning that fibers are stress free. Both cleaves’ quality is similar to
that of silica fibers.
Figure 10.9 A 9 μm core single-mode and a 450 μm core multimode

zirconium fluoride glass fiber.
10.4.2 Fiber Attenuation

Attenuation is without doubt the property of fluoride glass that
more attracts the attention of scientists and industrial communities.
In fact, when we compare attenuation between silica and fluoride
fibers, as shown in Fig. 10.10, we understand why reducing the
loss was the first technical race that took place between different
research groups. The theoretical attenuation is 2 orders of
magnitude lower for fluoride than for silica [42]. Tens of university
and industrial research groups across the world have participated
to develop this technology, including Centre National d’Etudes des
Telecommunications (CNET) and Alcatel in France, Corning and
Lucent and the Naval Research Laboratory (NRL) in USA, Nippon
Telegraph and Telephone (NTT) in Japan, and British Telecom in the
United Kingdom.
Unfortunately, after two decades of intense development, and
few tens of millions of dollars in funding, we are still far from the
goal. However, quite low loss has been reported by some groups.
Kanamori et al. from the NTT have reported 0.7 dB/km [43] and
Tran et al. from the NRL have reported 0.9 dB/km [44]. This loss is
a considerable achievement, considering the state of the technology
at that time, 1986, only eight years after the discovery of the first
fluoride glass. These values have been obtained on short lengths of

fibers and still 1 to 2 orders of magnitude higher than theoretical
values and even higher than already achieved attenuation for silica
fiber.
Figure 10.10 Comparison of theoretical loss of fluoride and silica fibers.
Total attenuation is due to intrinsic and extrinsic attenuations, as

shown by the equation below:
αTotal = αIntrinsic + αExtrinsic (10.4)
Extrinsic loss can be either absorption or scattering, due to
impurities, such as transition metals, rare earth elements, water,
microbubbles, crystals, complex anions such as sulfate and
phosphate [45], dissolved gases, and core cladding interface defects.
Intrinsic loss is due to the material matrix absorption and
scattering, including ultraviolet absorption, Rayleigh scattering, and
multiphonon absorption. Rayleigh scattering is due to density and
refractive index variation. Intrinsic attenuation can be written as
follows:
a Intrinsic = a UV + a Rayleigh + a Multiphonon
ÊA ˆ B Ê -C ˆ (10.5)
a Intrinsic = A1 exp Á 2 ˜ + 14 + C1 exp Á 2 ˜
Ë l¯ l Ë l ¯
Optical Fiber 327
Above 2 microns the ultraviolet absorption is negligible and the

loss is mainly due to Rayleigh scattering and multiphonon absorp-
tion. And above 3 microns, the Rayleigh scattering becomes negligi-
ble and the loss in mainly dominated by multiphonon absorption.
For more details, there are two interesting books dedicated to
fluoride glass and fiber attenuation measurement [46, 47].
The cut-back method is commonly used to measure fiber
attenuation. Using new Fourier transform infrared spectroscopy
(FTIR) from ABB we were able to measure attenuation for both
single-mode and multimode fibers. Of course the length of the fiber
to be measured is limited, less than 40 m. For longer fibers one need
a supercontinuum source that covers the transmission window of
the fiber. The alternative solution is to use, especially for single-
mode fibers, different lasers.
Current commercial fibers, including single-mode and multimode
fibers, have attenuation ranging from 50 to 100 dB/km. This is more
than enough for short- and medium-length applications. These fibers
are produced with very good yield, to be offered at a reasonable cost.
Figure 10.11 shows the attenuation curve of zirconium multi-
mode fiber from the ultraviolet up to 4.5 microns in the midinfra-
red, without any absorption peaks. The small absorption peaks at
a short wavelength are due to some transition metal impurities re-
maining in the starting materials. No –OH peak is visible on the spec-
trum around 2.9 microns, which means that the –OH concentration
is extremely low. In fact, the glass is prepared in a very controlled
environment. The –OH concentration inside the glove boxes is main-
tained lower than 0.1 ppm.
Figure 10.11 ZrF4 multimode fiber attenuation.

Indium fluoride glass fibers have a transmission window larger

than that of zirconium fluoride glass fibers. This is due to the fact
that indium fluoride has a lower phonon energy than zirconium
fluoride. Figure 10.12 shows a comparison of transmission curves of
indium and zirconium multimode fibers. Indium fluoride glass fibers
extend the application field of fluoride glasses and fibers.
Figure 10.12 ZrF4 and InF3 multimode fiber attenuation (comparison).
10.4.3 Mechanical Properties

Unlike optical properties of fluoride glass fibers, only limited work
has been reported on mechanical properties of exotic fibers. The
reasons are the cost of the fiber and the poor mechanical quality of
the fiber at the beginning of the development. One can’t draw any
good conclusion when using fiber that had many extrinsic defects,
and the preparation process is far from perfect. In fact the mechanical
quality of a fiber will depend on the mechanical properties of the
material itself and also on its preparation process, including glass
preparation, preform preparation, and the fiber-drawing process.
So, for mechanical studies, one has to take into account where the
fiber comes from, which process has been used to make the glass,
the preform preparation process, and the fiber-drawing conditions.
In fact, there is no standard process to produce exotic optical fibers.
There are as many processes as groups that were involved in the
development of these fibers.
The mechanical strength of all exotic optical fibers is without
doubt a big concern. Their practical use is ultimately linked to their
Optical Fiber 329
mechanical properties. In fact, most exotic fibers have low strength

and are far from being used in practice without problems.
Considering short- and medium-length applications, the solidity
of the fibers is more important than their optical loss. A fiber with
good strength and even only 95% transmission will be more useful
than a fiber with ultralow loss 99.9% transmission and that is
mechanically very weak.
Of course, there is no exotic fiber that can have a mechanical
strength comparable to silica fiber strength. This is due to the nature
of the material and its structure and the strength of interatomic
bonds.
As far as fluoride glass fibers are concerned, the theoretical
strength is about 50% to 60% lower than silica fiber strength. The
theoretical strength of fluoride glass fibers has been estimated to
be around 5 to 6 GP; this is between 700 and 800 kpsi [48, 49].
Theoretically, in practice one can expect obtaining fluoride glass
fibers with 40% to 50% the strength of current silica fibers, which
would be extraordinary. Unfortunately, we are still very far from this
reality. In fact, the preparation process of exotic glasses is still far
from perfect. However, significant progress has been made in this
area during the last two decades, and quite high-tensile-strength
fibers have been reported. Irphotonics, now Thorlabs, has reported
fibers with strength ranging from 75 to 130 kpsi and a bending
strength around 200 kpsi. This is the highest strength ever reported
for long-length midinfrared fibers. Figure 10.13 shows the Weibull
plot for tensile strength for a 200 micron InF3 multimode fiber. The
Weibull plot has been obtained using a 50 cm gauge, as described
in the FOTP 28 telecommunication standards. Figure 10.14 shows
the Weibull plot of two-point bending strength. Bending strength is
always higher than tensile strength because only a short length of
a fiber is submitted to stress during the test. There is still room for
improvement.
High mechanical strength makes the fibers easy to use and work
with. In fact fluoride glass fibers can be striped, cleaved, and spliced
using standard equipment. However, the equipment’s parameters
have to be adjusted.
ReliaSoft’s Weibull++ 6.0 - www.Weibull.com

Probability-Weibull
99.00
Weibull
Data 1
W2 RRX-SRMMED
90.00 F=12/S=0
50.00
Unreliability, F(t)
10.00
5.00
MSaad
Irphotonics
1.00 11/9/2011 09:17
10.00 100.00 1000.00
Stress at Breakage (kpsi)
b=4.1054, h=113.4186, r=00.9856
Figure 10.13 Weibull plot for InF3 multimode fiber tensile strength.
ReliaSoft’s Weibull ++ 6.0-www.Weibull.com

907018790 2 point bend test
99.00 Weibull
Data 1
90.00 W2 RRX-SRMMED
F=30 / S= 0
50.00
Unreliability, F(t)
10.00
5.00
Patrick Orsini
IR Photonics
2010-09-07 13:09
1.00
10.00 100.00 1000.00
Stress at breakage (kpsi)
b=5.8657, h=236.0039, r=0.9743
Figure 10.14 Two-point bending strength for InF3 multimode fiber.

Optical Fiber 331
10.4.4 Rare Earth–Doped Fibers

An optical fiber can be either a passive or an active medium. A passive
optical fiber will only transmit light from point A to point B without
changing its nature or its wavelength. An active fiber will change the
nature of light going through it under certain conditions. The fiber
has to be doped with active ions, such as rare earth elements. Fluoride
glasses are interesting and excellent hosts for rare earth elements.
In fact, rare earth ions can be part of the glass compositions. Stable
compositions contain more than 100,000 ppm of rare earth ions
without compromising glass stability. For comparison, the rare earth
concentration in silica glass fiber is only a few hundreds to a few
thousands of ppm according to the glass composition.
Active applications include fiber lasers, amplifiers, and laser
cooling. Tremendous work has been reported in these areas. Over 35
laser lines from the ultraviolet to the midinfrared region have been
reported using fluoride glass fibers doped or co-doped with different
rare earth ions. A very interesting and complete review has been
reported by Xiushan Zhu and N. Peyghambarian from the College
of Optical Science at Arizona University [50]. When this review was
written, the maximum output power reported for erbium-doped
fluoride fiber at 2.94 microns was 10 w, which was reported by
the Laval University team in Quebec City in Canada. Since then this
output has been tripled and a 30 W erbium laser at 2.9 microns has
been reported recently by the same team [51].
More recently, Quimby and Saad reported, according to lifetime
and spectroscopy measurement and theoretical modeling, that
efficient lasing by a 4.5 um fiber laser is possible using dysprosium-
doped indium fluoride glass fibers using a cascade lasing scheme
[52].
Laser cooling, which is the subject of this book, and will be deeply
addressed by experts, has been demonstrated using rare earth–
doped fluoride glasses and fibers. Most of the laser cooling work
has been done using ytterbium fluoride–doped fluoride glasses and
fibers [53–55]. However, thulium fluoride–doped fluoride glasses
have been also reported [56].
References
1. Yu, F., Wadsworth, W. J., Knight, J. C. (2012). Low loss silica hollow core
fibers for 3–4 μm spectral region, Opt. Express, 20(10), pp. 11153–
11158.
2. Poulain, M. (1995). New trends in halide glass compositions, J. Non-
Cryst. Solids, 184, pp. 103–108.
3. Conseil, C., Bastien, J.-C., Boussard-Plédel, C., Zhang, X.-H., Lucas, P., Dai,
S., Lucas, J., Bureau, B. (2012). Te-based chalcohalide glasses for far-
infrared optical fiber, Opt. Mater. Express, 2(11), pp. 1470–1477.
4. Wang, J. S., Vogel, M. E., Snitzer, E. (1994). Tellurite glass: a new
candidate for fiber devices, Opt. Mater., 3, pp. 177–203.
5. Guorong, C., Wen, L., Jijian, C. (2000). Improved properties for precision
molding phosphate glasses, Am. Ceram. Bull., pp. 44–46.
6. Feigelson, R. S., Kway, W. L., Route, R. K. (1984). Single crystal fiber by
the laser-heated pedestal growth method, Proc. SPIE, 0484 (Infrared
Optical Materials and Fibers III), p. 133.
7. Saad, M. (2009). Fluoride glass fiber: state of the art, Proc. SPIE, 7316,
p. 73160N.
8. Van Uitert, L. G., Wemple, S. H. (1978). ZnCl2 glass: a potential ultralow-
loss optical fiber materials, Appl. Phys. Lett., 33, pp. 57–58.
9. Poulain, M., Poulain, M., Lucas, J., Brun, P. (1975). Verres flores au
tetrafluorure de zirconium proprietes optiques d’un verre dope au
Nd3+, Mater. Res. Bull., 10, pp. 243–246.
10. Tregoat, D., Fonteneau, G., Lucas, J. (1985). Verres fluores a base de
metauxlours. Traitement des impuretes oxydes et hydroxides sous
atmosphere controlee, Mater. Res. Bull., 20, pp. 179–185.
11. Robinson, M. (1985). Preparation and purification of fluoride glass
starting materials, Mater. Sci. Forum, 5–6, pp. 19–33.
12. Tran, D. C., Sigel, G. H., Bendow, B. (1984). Heavy metal fluoride glasses
and fiber, a review, J. Lightwave Technol., LT2(5), pp. 566–586.
13. Nakai, T., Mimura, Y., Hokiwa, H., Shinbori, O. (1986). Dehydration of
fluoride by NF3 processing, J. Lightwave Technol., LT4, pp. 87–89.
14. Adan, J. L., Ricordel, C., Lucas, J. (1997). New composition of low
phonon energy fluoride and chloro-fluoride glasses, J. Non-Cryst.
Solids, 213–214, pp. 30–35.
15. Matecki, M., Poulain, Mi., Poulain, Ma. (1983). Cadmium halide glasses,
J. Non-Cryst. Solids, 56(1–3), pp 81–86.
References 333
16. Poulain, Mi., Matecki, J. M., Mouric, L., Poulain, Ma. (1983). Cadmium
halide glasses II. Chloride glass, Mater. Res. Bull., 18(5), pp. 631–636.
17. Matecki, M., Poulain, Ma. (1992). Composition adjustments in
fluorochloride glasses, J. Non-Cryst. Solids, 140, pp. 82–86.
18. Poulain, Mi., Poulain, Ma. Matecki, M. (1986). Rare-earths in fluoride
glasses: LnF3, AlF3, ThaF4, BaF2 systems and related glasses, J. Non-
Cryst. Solids, 51(2), pp. 201–215.
19. Bouaggad, A. (1986). PhD thesis, Rennes University, France.
20. Lecoq, A., Poulain, Ma. (1980). Etude phenomenologique du role
stabilisateur de l’aluminium dans les verres au tetrafluorure de
zirconium, Verres Refct., 34(3), pp. 333–342.
21. Jacoboni, C., Boulard, B., Baniel, P., Poignant, H. (1987). Vapor
deposition of fluoride glasses, Mater. Sci. Forum, 19–20, pp. 253–258.
22. Ohishi, Y., Mitachi, S., Kanamori T., Manabe, T. (1983). Optical absorption
of 3d transition metal and rare-earth elements in zirconium fluoride
glasses, Phys. Chem. Glasses, 24, pp. 135–140.
23. France, P. W. (1987). PhD thesis, Chapter 6, Sheffield University.
24. France, P. W., Carter, S. F., Williams, J. R., Beales, K. J., Parker, J. M. (1984).
OH-absorption in fluoride glass infrared fibers, Electron. Lett., 20(14),
pp. 607–608.
25. Drexhage, M. G. (1985). Preparation and properties of high quality
bulk fluoride glasses, Mat. Sci. Forum, 5, pp. 1–18.
26. Fujiura, K., Ohishi, Y., Sakaguchi, S., Ternuma, Y. (1989). Synthesis of
high-purity zirconium fluoride for fluoride glass fibers by chemical
vapor deposition, J. Am. Ceram. Soc., 72(8), pp. 1368–1372.
27. Withers, H. P. Jr., Monk, V. A., Cooper, G. A. (1989). High purity zirconium
tetra-fluoride for fluoride glass applications, Proc. SPIE, 1048, pp. 72–
77.
28. Saad, M. (2009). High purity fluoride glass synthesis: a review, Proc.
SPIE, 7228, p. 72280G.
29. Poulain, M. (1991). Chapter 1 Fluoride glass compositions and
processing, in Fluoride Glass Fiber Optics, Aggarwal, I., Lu, G., eds.
(Academic Press, Boston), pp. 1–35.
30. Fonteneau, G., Tregoat D., Lucas, J. (1985). Determination du Coefficient
d’Extinction de OH- dans les verres fluores a base de metaux lourds,
Mater. Res. Bull., 20(9), pp. 1047–1051.
31. Nakai, T., Mimura, Y., Tokiwa, H., Shinmori, O. (1986). Dehydration of
fluoride glasses by NF3 processing, J. Lightwave Technol., LT-4, 1.
32. Saad, M., Poulain, Ma. (1995). Sol-gel synthesis of fluorozirconate

glasses, Am. Ceram. Bull., 74(8), pp. 66–71.
33. Hruby, A. (1972). Evaluation of glass-forming tendency by means of
DTA, Czech J. Phys., B22, pp. 1187–1193.
34. Saad, M., Poulain, M. (1987). Glass forming ability criterion, Mater. Sci.
Forum, 19/20, pp. 11–18.
35. Uhlmann, D. R. (1983). Glass formation, a contemporary view, J. Am.
Ceram. Soc., 66, pp. 95–100.
36. Busse, L. E., Lu, G., Tran, D. C., Sigel, G. H. (1985). A combined DSC/
optical microscopy study of crystallization in fluorozirconate glasses
upon cooling from melt, Mater. Sci. Forum, 6, pp. 219–228.
37. Lipeng, Z., Fuxi, G., Wang, P. (1994). Evaluation of refractive-index and
material dispersion in fluoride glasses, Appl. Opt., 33(1), pp. 50–56.
38. Jiang, X., Joly, N. Y., Finger, M. A., Wong, G. K., Babic, F., Saad, M., Russell, P.
S. (2013). Close to three-octave-spanning supercontinuum generated
in ZBLAN photonic crystal fiber, Advance Solid-State Lasers Congress,
JTh5A, p. 6.
39. Tokiwa, H., Mimura, Y., Nakai, T., Shinbori, O. (1985). Fabrication of long
single mode and multimode fluoride-glass fibers by double crucible
technique, Electron. Lett., 21(24), pp. 1131–1132.
40. Mitachi, S., Miyashita, T., Kanamori, T. (1981). Fluoride glass cladded
optical fibers for mid-infrared rat transmission, Electron. Lett., 17, pp.
591–592.
41. Tran, D. C., Fisher, C. F., Sigel, G. H. (1982). Fluoride glass preforms
prepared by rotational casting process, Electron. Lett., 18(15), pp.
657–658.
42. Shibata, S., Horiguchi, M., Jinguchi, K., Mitachi, S., Kanamori, T., Manabe,
T. (1981). Prediction of loss minima in infrared optical fibers, Electron.
Lett., 17, pp. 775–777.
43. Kanamori, T., Sakaguchi, S. (1986). Preparation of elevated NA fluoride
optical fibers, Jpn. J. Appl. Phys., 25, pp. 468–470.
44. Tran, D. C., Levin, K. H., Burk, M. J., Fisher, C. F., Brower, D. (1986).
Preparation and properties of high optical quality IR transmitting
glasses and fibers based on metal fluorides, Proc. SPIE, 618, pp. 48–50.
45. Poulain, Ma., Saad, M. (1984). Absorption loss due to complex anions
in fluorozirconate glasses, J. Lightwave Technol., LT-2(5), pp. 599–602.
46. France, O. W., Drexhage, M. G., Parker, J. M., Moore, M. W., Carter, S. F.,
Wright, J. V. (1990). Fluoride Glass Optical Fibers (CRC Press).
References 335
47. Aggarwal, I. D., Lu, G. (1991). Fluoride Glass Fiber Optics (Academic
Press, Harcourt Brace Jovanovich, San Diego, CA).
48. Carter, S. F. (1990). Mechanical properties, in Fluoride Glass Optical
Fibers, P. W. France, ed., Blackie USA and Canada (CRC Press), p. 219.
49. Poulain, Ma., Gougeon, N., Maze, G. (2003). Mechanical properties of
fluoride glass fibers, Proc. SPIE, 4940, pp. 30–46.
50. Xiushan, Z., Peyghambarian, N. (2010). High-power ZBLAN glass fiber
lasers: review and prospect, Adv. OptoElectron., 2010, Article ID.
501956, 23 pages.
51. Fortin, V., Bernier, M., Bah, S. T., Vallee, R. (2015). 30W fluoride glass
all-fiber laser at 2.94 µm, Opt. Lett., 40(12), pp. 2882–2885.
52. Quimby, R., Saad, M. (2013). Dy-fluoroindate fiber laser at 4.5 μm with
cascade lasing, Adv. Solid State Lasers, paper AM2A.7.
Nature, 377, pp. 500–503.
54. Rayner, A., Friese, M. E. J., Truscott, A. G., Heckenberg, N. R., Rubinsztein-
Dunlop, H. (2001). Laser cooling of a solid from ambient temperature,
J. Mod. Opt., 48(1), pp. 103–114.
55. Murtagh, M. T., Sigel Jr., G. H., Fajardo, J. C., Edwards, B. C., Epstein,
R. I. (1999). Laser-induced fluorescent cooling of rare earth–doped
fluoride glasses, J. Non-Cryst. Solids, 253, pp. 50–57.
56. Hoyt, C. W., Hasselbeck, M. P., Sheik-Bahae, M. (2003). Advances in
laser cooling of thulium-doped glass, J. Opt. Soc. Am. B, 20(5), pp.
1066–1074.
Chapter 11
Crystal Growth of Fluoride Single

Crystals for Optical Refrigeration

Dipartimento di Fisica-NEST Instituto di nanoscienze CNR,
Universita`di Pisa, L. B. Pontecorvo 3, I-56127, Pisa, Italy
volpi@df.unipi.it
This chapter reports on the crystal growth of Yb-doped fluoride

single crystals for solid-state laser cooling. In the last years, the
advances enabled by Yb-doped fluorides have tremendously in-
creased the interest in this technology. Cryogenic operations have
been achieved as well as competitive steady-state cooling powers
in the temperature range below 250 K. Minimum achievable tem-
peratures are approaching 80 K. Truly vibration-free and extremely
compact, the optical cooling technology provides an ideal solution
for cooling down detectors, in space applications below the thermo-
electric cooler (TEC) barrier (170 K), where Peltier coolers cannot
operate and other techniques lead to complex solutions. The current
state-of-the-art results also prospect efficiency improvements in the
higher-temperature range, between 170 and 220 K, where the TEC
exhibits limited performances.

www.panstanford.com
338 Crystal Growth of Fluoride Single Crystals for Optical Refrigeration
Efficient cooling performances, however, are firstly related to the

optical quality of active materials. Stringent requirements on the
purity and structural quality need to be satisfied in order to achieve
net bulk cooling. Even though the purity of starting materials plays a
crucial role, the optical quality of finished crystals essentially relies
on the quality of the growth process. Low impurity concentrations
or isolated structural defects can be easily introduced during the
synthesis with tremendous effects on the efficiency of the anti-Stokes
process, which removes heat from the material. Careful techniques,
ultracontrolled conditions, and high-purity environments are basic
requirements to ensure the sufficient optical quality grade.
In our facility several Yb-doped fluoride crystals have been grown
for laser cooling experiments. Cryogenic operations were achieved
in 5 at% Yb-doped LiYF4 (YLF) single crystals in cooling experiments
carried out at the University of New Mexico [12, 17]. More recently,
varying Yb doping levels have been investigated and controlled co-
doped systems have resulted in efficiency enhancements via energy-
transfer-assisted anti-Stokes processes. The efficient performances
measured in cooling experiments reflect the high optical quality
of the grown samples. In the following sections, crystal growth
techniques and methods for assessment of crystal quality are
presented.
11.1 Introduction
In the solid phase thermal energy is largely contained in the
vibrational modes of the lattice. Upon interaction with laser light,
heat can be removed from the system through annihilation of lattice
phonons.
Optical cooling in solids is based on the principle of anti-Stokes
luminescence, where coherent excitation below the mean emission
energy of the electronic transition induces absorption of energy from
the lattice, and hence annihilation of lattice phonons, via ultrafast
thermalization of the electronic excitations (Fig. 11.1). Absorption of
photons at one frequency is followed by spontaneous re-emission of
blue-shifted photons. The extra energy is removed from the lattice,
resulting in bulk net cooling.
Introduction 339
λ>λf
λf
Figure 11.1 Schematic of anti-Stokes cooling. Following coherent

excitation in the low-energy tail of the absorption spectrum
(λ > λf), spontaneous emission of blue-shifted photons occurs
via annihilation of lattice phonons. Ultrafast thermalization of
the electronic excitations leads to absorption of energy from
the lattice through inelastic scattering with the vibrational
modes. The extra energy is removed from the lattice through
fluorescence of blue-shifted photons.
The idea of using optical radiation to lower thermal energy

from vibrational modes, through fluorescence upconversion, was
firstly proposed by Prighsheim in 1929 [16]. First observations,
however, required the invention of lasers. Anti-Stokes cooling
was firstly achieved in gaseous systems [6] and in 1995 the first
experimental demonstration in solid materials was accomplished.
At the Los Alamos National Laboratory, Epstein et al. observed net
cooling of a Yb3+-doped ZBLANP glass by optically pumping the 1
µm transition [7]. The prior invention of purification techniques
for fluorozirconate glasses was pivotal for such achievement,
enabling the synthesis of high-purity active materials. Subsequent
optimization of experimental conditions led to optically cool a highly
purified Yb-doped ZBLAN glass to 208 K from room temperature
[18].
Dramatic improvements of cooling performances, however, have

been achieved only recently with high-quality fluoride single crystals.
Conversely to glassy matrices, in crystalline hosts the long-range
order leads to sharp homogeneous splitting of rare earth energy,
levels resulting in dramatic enhancement of resonant absorption.
Fluoride crystals also possess a low phonon energy, which strongly
reduces nonradiative decays rates, leading to high radiative
efficiencies. Additionally, crystalline hosts stoichiometrically allows
for a high doping level and thus the resonant absorption of the
electronic transition can be largely enhanced by just increasing the
concentration of active ions.
High-purity Yb-doped fluoride single crystals have definitely led
to cooling operations in the cryogenic regime. The first breakthrough
result was achieved in 2010 with the optical cooling of a 5 at%
Yb-doped LiYF4 (YLF) single crystals down to 155 K from room
temperature [17]. Optimization of the pumping scheme led to the
first operation below the NIST-defined cryogenic temperature (123
K) [12]. The increase of Yb doping level up to 10 at% in ultrapurified
YLF single crystals enabled the first optical cooling below 100
K [14]. Current operations, however, are limited by impurity-
mediated background absorption. Calculations indicate that further
purification of starting materials would enable operations close to
the liquid nitrogen temperature (77 K) [13]. Due to the chemical
inertness of binary fluoride, however, purification of raw materials
below the part-per-million level is still a quite challenging process.
One approach currently under investigation is detailed in Ref. [10].
Cryogenic achievements have dramatically increased the
technological interest in this field. Current operations and cooling
powers available in the temperature range of 90–250 K definitely
allow one to consider in earnest device implementation. Truly
vibration free, such a solution is much sought after for space
applications. The inherent absence of vibration and stunning
compactness, indeed, would permit significant improvements for
several high-precision space-based technologies, including focal
plane infrared (IR) imaging detectors, ultrastable cavity lasers for
atomic clocks and gravitation wave detection, and X-ray and gamma
ray sensors. In the temperature range of 80–150 K solid-state optical
coolers would also enable the use of superconductor devices.
Introduction 341
With the advances achieved in cooling experiments, the

investigation in the synthesis of more and more pure and defect-
free fluoride crystals has attracted enormous interest as well as the
investigation of purification techniques for raw materials. Efficient
cooling performances are primarily related to the quality of active
materials, and both device implementation and fundamental
research require high-quality materials for efficient operations and
further improvements of the current state of the art, pushing down
the minimum achievable temperature close to the thermodynamic
limit.
High optical purity and structural quality are basic conditions
for efficient bulk cooling. Impurity-mediated processes and
crystallographic defects introduce severe background absorption,
thereby lowering the anti-Stokes efficiency. Spurious electronic
and vibrational bands cause trapping of pump photons as well as
nonradiative quenching of the electronic excitations via energy
transfer and multiphonon relaxation. Besides degrading the
absorption efficiency, such processes lower the external quantum
efficiency of the electronic transition and generate heat dissipation
via multiphonon emission. Maximizing the cooling efficiency firstly
requires minimizing impurity concentrations. High structural
quality is mandatory as well. Crystal defects, such as microbubbles,
dislocations, and cracks, cause strong losses of pump power through
scattering processes of pumping photons, violently heating up the
lattice.
In addition to degrade the anti-Stokes efficiency, all these
processes generate detrimental heating. Such heat dissipation can
readily compromise the cooling process if the efficiency of heat-
generating processes is not surpassed by the efficiency of the anti-
Stokes process, which removes heat from the lattice. As shown in
several experiments, even sparse distribution of defects and very
low concentration of impurities heavily affect, and even compromise,
the cooling performances [5, 13]. Stringent requirements on optical
purity and structural quality of active materials come after.
Although a minimum concentration of impurities is imposed
by the purity grade of starting materials, the absence of structural
defects and polycrystalline nuclei as well as the inclusion of
additional impurities are exclusively related to the quality of the
synthesis procedure. Growth is a very critical process for the optical
quality of finished materials. Impurities can be easily incorporated

during the synthesis and defects can readily form due slightly
unsuitable thermal and kinetic growth conditions. As shown in the
following, ultra-accurate procedures, controlled conditions and
highly purified environments are basic requirements to achieve the
sufficient optical quality grade.
Fluoride single crystals possess exceptional physical properties
as hosting materials for anti-Stokes cooling. The growth of high-
purity samples, however, requires specific conditions due to the
strong reactivity of fluorine. The main advantage of these materials,
over other crystalline hosts, is the low cutoff energy (300–500 cm–1)
of optical phonons that effectively suppresses the efficiency of pho-
non-mediated relaxation processes, leading to high radiative quan-
tum efficiencies. The lower phonon energy of bromides, chlorides,
and iodides does not offer any appreciable advantage if compared
to the complexity associated with the extreme hygroscopicity typi-
cal of heavy halides. Fluoride crystals are instead not hygroscopic.
Additionally, they have the advantage of low refractive index and
good thermal and mechanical properties. The low refractive index
reduces photon trapping due to internal reflections at the surfaces,
while good thermal conductivity enables efficient heat removal, and
the relatively high mechanical hardness lends benefits to sample
processing and surface polishing. Largely used for years as gain me-
dia for efficient solid-state lasers even in commercial devices, rare
earth–doped fluorides are currently under investigation for a variety
of novel applications, such as upconversion efficiency enhancement
in solar cells [8, 9], quantum computation, and particle detection.
Due to the strong reactivity of fluorine compounds with
moisture, however, the growth process must be carried out
under ultracontrolled conditions in order to avoid detrimental
contaminations. The most difficult impurities to avoid when growing
fluorides are oxygen complexes, such as OH– hydroxyl radicals,
which efficiently substitute for F– ions in the crystalline structure.
Such complexes, when incorporated inside crystals during the
growth, strongly degrade the overall optical performances due to
spurious absorption bands in the ultraviolet (UV) and near-infrared
(NIR), which act as efficient traps for electronic excitations in energy
transfer and multiphonon relaxation processes. Especially in the
case of Yb3+ ions, inclusion of OH– ions makes nonradiative decay
The Czochralski Growth Method 343
the dominant relaxation process from the excited state, dramatically

lowering the radiative efficiency of the electronic excitation.
The growth of fluorides doped with Yb3+ ions imposes further
conditions, increasing the complexity of the synthesis process. Due to
the tendency of Yb3+ ions to reduce to the 2+ state, a suitable oxidant
atmosphere need to be used in order to prevent any inclusion of
Yb2+ impurities inside growing crystals, which form from reduction
of Yb3+ ions. In addition to suppress the efficiency of cooling process,
reducing the number of Yb3+ active ions, the inclusion of Yb2+ ions
inside growing crystals significantly deteriorates the optical quality.
Such ions indeed possess large absorption bands in the visible region,
which are likely to participate in nonradiative transfer of excitations
from Yb3+ ions, preventing anti-Stokes emission. Fluorinating agents,
however, can lead to other kind of contaminations and hence require
proper use.
In the following section, methods and growth details are
presented for different Yb-doped samples grown in our facility
for laser cooling applications. Methods used for optical quality
assessment after the growth are also discussed.
11.2 The Czochralski Growth Method

The Czochralski technique is nowadays most commonly used for the
synthesis of bulk semiconductors, oxide and fluoride single crystals
and synthetic gemstones for photonic applications. Originally
invented in 1918 by Czochralski [4], this process is capable of
fabricate defect-free single crystals of high optical quality, with a
large size, at a relatively fast growth rate [2, 15].
Crystal growth is involved with the control of a phase transition.
Growth methods from melt, such as the Czochralski process, are
based on the control of the solid–liquid-phase transition to induce
the crystallization. The liquid is slightly undercooled below its
freezing point and the nucleation of a single-crystal domain is
induced by using a crystal seed with the same crystal structure and
composition of the material to be grown. A principle diagram is
shown in Fig. 11.2.
Raw materials are normally stored in powder form. The mixture
of starting materials is first melted inside a crucible and then
brought in an undercooled phase, lowering the temperature of the

liquid. In the undercooled phase, the liquid tends to crystallize. The
crystallization is induced at the solid–liquid interface by using a
crystal seed, dipped into the melt. Once the nucleation occurs, the
growth starts. The crystal is pulled from the melt slowly moving
upward the crystal seed, while it is rotated, at the same time, in
order to ensure thermal symmetry, hence preventing the growing
crystal to assume an irregular cross-sectional shape related to the
temperature distribution in the melt.
Fundamental condition for the Czochralski growth is that the
temperature of the solid–liquid interface must be the lowest in the
liquid. If it is not, spurious nucleation can occur.
Figure 11.2 Schematic of Czochralski growth.
The basic requirement to grow defect-free single crystals is the

use of a slow pulling rate. Fast growth speeds commonly give rise
The Czochralski Growth Method 345
to inclusion of dislocations, nucleation of polycrystalline domains,

and lack of transparency. For fluorides typical pulling rates vary
between some mm/h and tens of mm/h. The optimal growth rate
depends on the physical and thermal properties of the material.
A good compromise between growing as fast as possible and the
introduction of defects that this produces is normally chosen.
Rotation rates of a few revolutions/min are normally sufficient to
achieve defect-free samples.
One of the main advantages of the Czochralski method is the
possibility to control the diameter of the crystal while it is growing.
The crystal diameter at the solid–liquid interface depends on both
the temperature of the melt and the crystal pull rate and can thus
be controlled by changing both these parameters. Decreasing
the temperature of the melt or the pulling rate leads to diameter
enlargements, while the opposite is needed for shrinking the
diameter. To minimize formation of defects, however, the diameter
control is commonly run by changing the melt temperature, while
maintaining fixed the pulling rate.
Ultra-accurate control of the melt temperature is fundamental
condition for the growth process. Irregular temperature changes
reflect in sudden diameter variations. In proximity of such diameter
changes, clouds of defects typically form due to fast growth rates in
certain directions. Temperature instabilities also lead to spurious
nucleation in the melt. Controlled diameter changes and uniform
growth at sufficient slow rates require fine-tuning and high stability
of the melt temperature. Suitable heating systems thus need to be
employed.
A key point for the Czochralski growth process is the quality
of the crystal seed, which defines the crystalline structure and the
orientation of the unit cell in the nucleation process. An ideal seed
would be a defect-free oriented single crystal of the same composition
of the material to be grown. However, the basic requirements
to be satisfied are the same crystal structure and similar lattice
parameters. Especially for the growth of novel materials, if crystal
seeds are not available, a wire or a tube, properly shaped, can be
used to induce the nucleation of a polycrystalline domain, from
which single crystals can be obtained, alternatively increasing and
decreasing the diameter.
A further advantage of the Czochralski technique is the possibility

to carry out a necking operation which both enables prevention of
the propagation of nonaxial defects from the seed to the growing
crystal and favors the propagation of a single-crystal domain in the
growing crystal.
Such a procedure simply consists of making the crystal diameter
as small as possible by increasing the melt temperature. Shrinking
the crystal diameter either favors the propagation of a single-
crystal domain, reducing the number of crystallites propagated into
the crystal, or allows one to interrupt the propagation of nonaxial
dislocations within the growing crystals (see Fig. 11.3). When a
sufficient length of narrow crystal has been grown, the diameter
can be enlarged to the desired value. The necking process can be
repeated several times, if necessary.
Figure 11.3 Schematic of necking technique.
Although a vacuum chamber would be the purest environment as

regard contaminations, the Czochralski process is commonly carried
out in controlled atmosphere as the molten material cannot be stable
at low pressures. To minimize instabilities which typically lead to
evaporation of material from the melt surface, the growth chamber
is usually filled, at proper pressure, with a purified inert gas. The
kind of gases constituting the growth atmosphere depends on the
materials to be grown. Argon is normally chosen for the growth of
fluoride crystals, while tetrafluoromethane (CF4) is typically used to
build up an oxidant atmosphere.
Crystal Growth of Fluorides 347
The basic requirement to avoid inclusion of contaminants during

the growth process is to keep the melt surface completely free from
foreign particles. Impurities deposited on the melt surface can
readily get incorporated inside growing crystals, compromising the
optical quality of samples. In addition to introducing loss channels
through spurious absorption bands, foreign particles at the growth
interface can originate irregular anisotropic growth rates, leading to
dislocations. Spurious nucleations on the melt surface are another
typical result of impurities deposited on the melt surface. Careful
cleaning operations must be performed before start of the growth.
The maximum sizes that can be produced by the Czochralski
technique are only related to the dimensions of the furnace. Basically,
the volume of the crucible and the distance that the seed holder can
move determine the maximum crystal dimensions. Typical sizes of
crystal boules used for optical applications are of the centimeter
scale.
Basic components of a Czochralski furnace are a growth chamber
and a pulling system. The pulling of the crystal from the melt takes
place inside the growth chamber, where are placed the heater
element and the crucible containing the raw materials. The pulling
system is designed to control the rototranslation movements of the
seed holder shaft, required either to bring the crystal seed in contact
with the melt or to pull the growing crystal from the melt surface. A
detailed description of the furnace, with all its components, designed
at the NMLA Laboratory for the growth of fluorides single crystals is
outlined in Section 11.4.
11.3 Crystal Growth of Fluorides

Due to the strong electronegativity of F– ions, for the growth of
fluorine-based crystalline compounds, specific requirements need
to be met in order to achieve high optical quality. Oxygen complexes,
which form when oxygen and water molecules are present during
the growth process, are the most detrimental impurities to avoid
when growing fluorides. Among these complexes, hydroxyl radicals,
OH–, are the most favorite to form as they efficiently substitute
for F– ions in fluoride crystalline structures. These ions exhibit
strong UV electronic absorption and IR vibrational bands [3],
which cause either trapping of pumping radiation or luminescence

quenching of electronic excitations. Spurious IR bands, however,
are the most detrimental for the efficiency of the cooling process
preventing anti-Stokes fluorescence via multiphonon relaxation,
that is, heat generation. In particular, IR absorption traps due to
hydroxyl radicals (around 3400 cm–1) cause significant multiphonon
quenching even for large-gap (>4000 cm–1) electronic transition,
which possess a negligible nonradiative decay rate in pure fluoride
crystals. This is the case of Yb-doped materials, where the presence
of OH– contaminants dramatically reduces the radiative efficiency
of the electronic transition. In such cases anti-Stokes fluorescence
is hindered and heat is generated via multiphonon emission.
Depending on the impurities’ concentration, the cooling process can
be even compromised due to a faulty balance between heating and
virtuous cooling processes.
The basic condition to prevent formation of OH– radicals in the
melt is to avoid interaction of raw materials with oxygen and water
at high temperatures. Specific conditions on the growth conditions
and raw powders’ storage and handling follow.
An essential requirement for furnaces dedicated to the growth
of fluorides is a high-vacuum system which enables the evacuation
of the growth chamber at low pressure, removing any presence of
oxygen and water molecules. Pressures of the order of 10–5 Pa are
normally sufficient for purification of the growth environments.
Storage of raw powders and preparation of the starting mixture
require proper procedures as well. Before loading into the furnace,
the fluorine raw materials must be stored inside sealed boxes.
Weighting and mixing of primary components inside the crucible
must be carried out inside a moisture-controlled environment.
To prevent instabilities of the molten material at high
temperatures, the growth process is typically carried out under inert
atmosphere. After high-vacuum cleaning, high-purity (5N) argon is
pumped in overpressure inside the growth chamber. For the growth
of fluorides doped with rare earth ions that tend to reduce from the
triply ionized state to the 2+ state, such as Yb3+, high-purity (5N)
tetrafluoromethane (CF4) is typically used as fluorinating agent. A
proper percentage of CF4 is pumped inside the furnace together with
argon. The dissociation of CF4 at high temperatures releases F– ions,
Crystal Growth of Fluorides 349
which create an oxidant environment, preventing triply charged Yb

ions to reduce to double-charged states.
The composition of the growth atmosphere is a critical parameter
for the optical quality of Yb-doped crystals and requires a previous
dedicated investigation. A lack of CF4 concentration in the growth
atmosphere leads to the formation of detrimental Yb2+ impurities
in the melt. An excess of CF4 concentration has multiple drawbacks
as well. Firstly, it likely leads to carbon contamination inside the
melt. Additionally, a CF4-enriched atmosphere is rather corrosive
for furnace components, especially in the case of extended use. The
deposition of a dark scum inside the furnace inside the furnace is
the most common result of an excess of CF4 in the atmosphere
composition.
Melting temperatures of fluoride materials are normally below
1000°C. Resistive heating systems are the most commonly used
for furnaces dedicated to the growth of these materials. Maximum
temperatures achievable are of the order of 1500°C. In resistive
systems, heat is generated via the Joule effect by the passage of
a current through a conductor surrounding the crucible, which
contains the raw materials. High stability and fine-tuning of the melt
temperature can be achieved. A sufficiently low vertical thermal
gradient can be attained, thanks to the high uniformity of the thermal
field generated by the resistive element.
For the furnace components, materials resistant to the corrosive
atmosphere at high temperatures must be used. Usual materials
employed are high-purity graphite and platinum. The former is
suitable to manufacture the heater element. Graphite combines
conductive behavior and excellent electrical resistivity for Joule
heating at high temperatures. It can operate up to 2500°C in an
inert atmosphere. Moreover, compared to most of heating element
materials, it can be readily machined in self-supporting tubular
shapes, suitable to surround the crucible and thus providing uniform
heating inside.
Regarding crucibles, the basic requirement is to avoid any
contaminations of raw materials at high temperatures. The crucible
thus must be made of high-purity materials which possess high
stability under the temperature and pressure conditions of growth
and hence do not interact with the material to be grown even at
high temperatures. Low-porosity materials are preferable. Among
materials chemically and thermally compatible to the fluorides’

growth conditions, platinum, graphite, and glassy carbon are more
suitable, as they additionally are not wetted by the fluoride melting.
Platinum is, however, the best choice as regard to purity. Relatively
chemically inactive, it is highly pure and has advantages of excellent
resistance to corrosion and stability at high temperatures. Its melting
temperature is about 1770°C. Being malleable and ductile, it can be
readily hammered into thin sheets or drawn into thin wires. Such
properties make high-purity platinum the best choice not only for
crucibles but for all the components that directly contact the melt,
such as seed holders, wires to fasten the crystal seed, and cleaning
tools.
While the melting point is reached, foreign particles tend to
condense above the melt surface, producing a dark, thin film.
Especially in the case of a CF4-enriched atmosphere, the extended use
of a corrosive atmosphere tends to increase the porosity of graphite
components, and the foreign particles, which are released when the
furnace is heated, tend to get absorbed more by the pores. Suitable
cleaning techniques must be employed before start the growth.
Clean surfaces, free from foreign particles, appear completely clear.
A typical technique used for the growth of fluorides under a CF4-
enriched atmosphere consists in freezing the melt surface by using
a crystal seed, which is lowered in contact to melt so as to induce
the formation of a large and thin crystal, collecting the impurities
deposited above the surface.
After the growth, the crystal boules need to be slowly cooled
to room temperature before be extracted from the furnace. Slow
cooling rates need to be used in order to prevent thermal shock
which can cause internal cracks or phase changes. Typical cooling
rates are of the order of tens of Celsius degrees per hour.
11.4 The Growth Facility at the NMLA

Laboratory of Pisa University
The growth facility for fluoride single crystals at the NMLA
Laboratory of Pisa University consists of two home-made Czochralski
furnaces, properly conceived for the growth of this kind of materials.
The growth of bulk cooling materials is carried out in a dedicated
The Growth Facility at the NMLA Laboratory of Pisa University 351
furnace, specifically designed for the growth of fluorides under CF4-

enriched atmosphere. The furnace is equipped with a high-vacuum
system, resistive heating and computer-controlled optical diameter
control. A schematic of the apparatus is shown in Fig. 11.4, along
with a photograph of its exterior.
Figure 11.4 Schematic (a) and picture (b) of the Czochralski furnace at the
NMLA Laboratory of Pisa University.
The main body of the furnace is composed of a primary growth

chamber connected by a gate valve to a secondary vacuum cell
placed above. Inside the growth chamber are produced the high
temperatures and pressure conditions necessary for the crystal
growth.
The secondary cell has been designed to enable access to the
bottom end of the puller shaft without altering the temperature
and pressure conditions of the growth chamber. This possibility is
crucial for the cleaning operation. Once the gate valve is closed, the
access to the bottom end of the puller allows to repeatedly lower
the cleaning tools into the melt, while maintaining completely
unchanged temperature and pressure conditions of the melt surface
between subsequent plunges.
Cooling to room temperature the molten material and degradation

of the growth atmosphere would be otherwise necessary to replace
the cleaning tools after being plunged into the melt, with consequent
poor efficiency of the cleaning operation. Additionally, such a design
is crucial to start the growth process by lowering the seed in contact
with the melt, without altering the conditions of the melt surface,
once the cleaning operation is completed.
The growth chamber below is constituted of a stainless steel
cylindrical body with a removable spherical cap. By removing the cap,
the crucible can be loaded inside the kiln. On the top of the cap, three
circular windows enable monitoring of the interior of the chamber
and hence the growth process. The entire exterior of the chamber is
water-chilled through a tight coil of copper pipes, soldered on the
external surface, within which cold water flows. Inside the chamber
are placed the heater element and the crucible containing the raw
materials to be melt. The heater element consists of a graphite
resistance with cylindrical geometry, which accommodates the
crucible in the middle. Such geometry has been properly conceived
to achieve uniform heating of the crucible and hence a uniform
temperature in the melt. By changing the vertical position of the
crucible within the height of the resistor, the proper thermal gradient
at the interface can be selected. Below the resistance are placed the
two electrodes. A schematic of the mounting is shown in Fig. 11.5.
A central hole on the bottom of the crucible holder enables
the connection with a thermocouple, by which is measured the
temperature inside the growth chamber. Outside the furnace, the
thermocouple is connected to a digital multimeter, which returns
the temperature value measured on the bottom of the crucible.
Beneath the crucible holder, a flange, by means of a feedthrough,
enables moving the crucible within the structure of the resistor.
Two thermal shields are positioned around the resistor to limit heat
leakage outward.
The temperature control inside the furnace is operated by
means of a PID (proportional–integral–derivative) control, which
automatically adjusts the output current of the power supply
accordingly to the discrepancy between measured and set values
in the feedback loop. Fine temperature changes of the order of few
tenths of Celsius degree can be achieved. The temperature stability
in growth conditions is less than tenth of a Celsius degree.
The Growth Facility at the NMLA Laboratory of Pisa University 353
(a) (b)
Figure 11.5 (a) Interior of the growth chamber. The heater element, with
cylindrical geometry, accommodates the crucible holder in the
middle. Beneath the resistance are placed the two electrodes.
(b) Platinum crucible used for the growth of Yb-doped YLF
single crystal.
A high-vacuum system enables evacuation at low pressure

both the growth chamber and the secondary cell. The apparatus is
comprised of two sequential vacuum pumps, connected separately
to the two chambers. A prevacuum is operated through a mechanical
rotary pump, reaching a pressure of 10–1 Pa. A turbomolecular pump
is sequentially employed to operate high vacuum, up to a level of
10–6 Pa.
A suitable system is installed outside the furnace to pump inside
the growth chamber controlled concentrations of high-purity (5N)
Ar and CF4 gases. The apparatus is designed to separately fill in Ar
and CF4. The gas amount pumped inside the chamber is controlled
by a pressure gauge installed at the entrance valve with 102 Pa
sensitivity.
The pulling system, which produces and controls the
rototranslation movement of the seed holder shaft, is installed above
the main body of the furnace (see Fig. 11.4). Basically, the apparatus
consists of a stainless steel puller shaft, controlled by two stepper
electrical motors, one for vertical translation and one for revolution.
The mechanical design ensure steady, fine and smooth movements
of the seed holder shaft, essential to avoid thermal shock and
irregularities of the growth rates with consequent degradation of
the structural quality of growing crystals. The arrangement ensures
the bottom end of the rod to move back and forth from the secondary
chamber to the bottom of the crucible. Minimum displacements
achievable are of the order of the tenth of a millimeter. The slowest
translational speed is 0.1 mm/h. The stepper motor that produces
the revolution of the shaft enables rotation rates of the order of a few
revolutions/min, normally required to attain thermal symmetry.
The puller shaft is terminated by a dedicated extension (seed
holder), which consists of a cylindrical tube, made of high-purity
platinum (99.95%). On the bottom end of the seed holder extension,
crystal seeds and the cleaning tools are fastened by using high-purity
platinum wires (99.95%).
The diameter of the growing crystal is controlled by using an
optical system computer controlled (see Fig. 11.6).
Figure 11.6 Schematic of the optical diameter control system.
The apparatus constitutes a red diode laser, which shines the

solid–liquid interface meniscus, and a charge-coupled device (CCD)
camera, which acquires the bright point due to the reflection of the
laser beam by the meniscus. The laser beam is aligned to produce
Crystal Growth of Cooling Materials at the NMLA Laboratory 355
an asymmetric spot at the interface so that the laser beam can shine
the meniscus while it is moving toward the crucible edge when
the crystal diameter increases. The reflection position acquired in
a revolution period is converted in a diameter value by means of
a suitable algorithm. The laser and the CCD camera are arranged
inside a compact support placed outside the growth chamber, in
proximity of one of the top window of the main body, through which
the incident beam is passed and the reflection from the meniscus
is acquired. The diameter control is operated by means of a second
PID feedback loop, which automatically varies the current supplied
to the resistance, and hence the melt temperature, according to the
difference between measured and set values of the crystal diameter.
Diameter fluctuations attainable in growth conditions are of the
order of 1%.
11.5 Crystal Growth of Cooling Materials at the

nmla Laboratory: Recent Results
Several Yb-doped fluorides have been grown in our facility for
laser cooling experiments. So far, major results in bulk cooling have
been achieved with the growth of high-quality YLF single crystals.
Cryogenic operations were achieved in 5 at% Yb-doped YLF samples
in cooling experiments carried out at the University of New Mexico
[1, 2]. More recently, YLF single crystals with increasing Yb doping
levels, up to 10 at%, have been grown, resulting in efficiency increase
via significant background absorption diminishing. A novel scheme
for Yb efficiency enhancement, based on energy transfer processes,
has been demonstrated with the growth of Yb–Tm co-doped YLF
single crystals. Remarkable cooling efficiencies were measured in a
single-pass configuration with both single Yb-doped and Yb–Tm co-
doped samples [21].
The high optical quality achieved in the YLF samples, and
verified in laser cooling experiments, is the outcome of a thorough
investigation carried out to identify the optimal growth parameters
while assessing the crystal quality after the growth. In addition to the
purity grade of the starting materials, specific growth parameters,
that is, composition of the growth atmosphere, thermal gradients,
choice of growth axis, melt composition, and pulling rate, critically
affect the crystal quality and hence need to be properly investigated

to achieve single crystals with a sufficient quality grade. After the
growth, the optical and structural quality of finished samples needs
to be assessed. Inclusion of internal defects, polycrystalline nuclei,
and major contaminants needs to be investigated with suitable
techniques.
This section focuses on these topics. Growth details and crystal
quality assessments are discussed for the Yb-doped YLF single
crystals grown in our facility.
11.5.1 The YLF Crystal

The YLF crystal is a birefringent insulating material largely used
as a laser-active medium, even in commercial devices. The crystal
possesses a band gap of about 7 eV and is optically transparent from
the UV (180 nm) to the IR (6.7 µm), enabling its use as a crystalline
host for almost all the trivalent rare earth ions. It is particularly
appreciated for its low indices of refraction and negative thermal
lensing. Additionally, among fluorides, it possesses a relative high
thermal conductivity and mechanical hardness. The crystal structure
is tetragonal with lattice parameter a = 5.197 Å and c = 10.735 Å and
the unit cell contains four molecules inside. A scheme of the primitive
cell of the YLF crystal is reported in Fig. 11.7. The crystallographic
space symmetry group is I41/a (Hermann–Mauguin notation). The
lattice type is body centered (I) and the crystal structure possess a
screw axis (41) and a glide plane (/a) symmetries.
The unit cell is thus invariant under a rotation of 90° (360°/4)
followed by a translation of half the lattice vector (41 screw axis
symmetry) and for a combination of a translation of half the lattice
vector along an a axis and a reflection in that plane (/a glide plane
symmetry). Crystals belonging to the tetragonal class are optically
uniaxial, which means they possess a unique direction along which
the two allowed orthogonal polarizations undergo the same phase
velocity. Such direction corresponds to the crystallographic c axis in
YLF crystals and defines the optical axis z of the crystal. The other
two optical axes x and y possess the same refractive index (nx = ny
≠ nz) and coincide with the two equivalent crystallographic axes a
of the unit cell. Only light traveling along the c axis is not subjected
to birefringence inside the crystal. Trivalent RE ions substitutionally
enter the Y3+ sites with point symmetry S4 and coordination number
8. Charge compensation is not required and the replacement of Y3+

with RE3+ ions does not significantly affect the crystal structure due
to the similarity of ionic radii sizes. Slight discrepancies in ionic
radii, however, can cause segregation phenomena. In the case of Yb
doping, the segregation coefficient is expected to be equal to 1 as the
ionic radii of Y3+ and Yb3+ for coordination number 8 in fluorides
are 1.16 Å and 1.12 Å, respectively [11]. A detailed list of optical and
physical properties of the YLF crystal is reported in Table 11.1.
Figure 11.7 Scheme of the unit cell of YLF crystal.
Table 11.1 Main physical and optical properties of the YLF crystal
Thermal conductivity @ 298 K (W/mK) 5.2 (a axis)

7.2 (c axis)
Transparency range (µm) 0.18–6.7
Refractive index (@ 1 µm) no = 1.45
ne = 1.47
Phonon energy (cm–1) 450
Density (g/cm3) 3.99
Specific heat (J/gK) 0.79
Coefficient of thermal expansion (10–6 K)–1 8.3 (a axis)
13.3 (c axis)
Hardness (Knoop) 300
11.5.2 The Growth of Yb-Doped YLF Single Crystals
11.5.2.1 The phase diagram of the LiF–YF3 system

Detailed knowledge of phase diagram is the starting point for
any growth process. The phase diagram provides a temperature
composition map which represents the thermodynamic phases of
the system as a function of temperature and relative concentration
of components. Temperature and composition conditions suitable to
achieve the growth along the phase transition between a single solid
phase and homogeneous liquid need to be identified as well as single
equilibrium phases and temperature-dependent phase transitions
relative to the compound composition.
The phase diagram relative to the YLF compound is shown in
Fig. 11.8 [19]. Such diagram represents all the equilibrium phases
and transition lines of the binary LiF-YF3 system as a function of
temperature and relative concentrations of lithium fluoride (LiF)
and yttrium fluoride (YF3), which are the constituents of the YLF
crystal. On the left and right sides of the diagram are the behaviors
of the pure LiF and YF3, respectively. For the YLF crystal, the
phase diagram indicates a strong incongruent melting behavior,
which means that the YLF crystal does not melt uniformly through
formation of liquid LiF and YF3 but, instead, increasing temperature,
LiF fuses firstly, while YF3 stays solid (YLFÆ LiF + solid YF3). A
displacement from the stoichiometric composition is thus required
in order to intersect the transition line between the YLF solid phase
and the homogeneous liquid phase, along which the growth occurs.
The peritectic single point occurs at 49 mol% YF3, the eutectic point
occur at 19 mol% YF3. An ideal phase for the undercooling state
is that with the coexisting YLF crystal and homogenous liquid (B),
which extend from 19 mol% to 49 mol% YF3. In principle this whole
concentration range would be suitable for growth. However, to
maximize the temperature extension for the growth process and to
keep a ratio between the components close as possible to 1, a high
concentration of YF3 is typically chosen.
To grow YLF single crystals with a given doping level of rare
earth3+ ions, a partial amount of Y3+ ions need to be replaced by RE3+
ions. A suitable concentration of YF3 hence need to be removed from
the starting mixture and replaced with the dopant compound REF3.
In case of co-doping with two different species of RE3+ ions, the YF3
compound needs to be replaced with appropriate amounts of both
the RE compounds. The dopant RE3+ ions, present in the melt, will be
incorporated in the crystal while it is growing, replacing the Y3+ host
ions in the unit cells.
Liquid
YF3 + L
T [K]
A B
LiF + YLF YF3 + YLF
Figure 11.8 Phase diagram of the LiF-YF3 system. The label A indicates a
phase with solid LiF mixed with liquid. The label B indicates a
phase with solid YLF and liquid [19].
The final composition in the case of single doping can be

expressed as: Li(Y1–xREx)F4, where x indicates the molar percentage
of RE3+ ions. In case of co-doping with two species of rare earth
ions RE13+ RE23+ the final composition can be expressed as: Li(Y1–x–
3+
yRE1xRE2y)F4 where x and y indicates the molar percentage of RE1
3+
and RE2 ions, respectively.
11.5.2.2 The growth of YLF single crystals with varying Yb

doping levels
The use of high-purity raw materials is the starting point for the
growth of efficient cooling crystals. Yb-doped YLF single crystals for
laser cooling were grown starting from fluorine raw powders of LiF,
YF3, and YbF3 with guaranteed 5N purity (99.999%). Such a purity
grade ensures that the sum of impurities contained in the binary
compounds is less than 10 ppm. Very few companies in the world
can supply such high-purity fluorine powders. Raw materials used

in this work where provided by AC Materials (Tarpon Spring, FL,
USA), a world-class company in this field, specialized in preparation
of high-purity binary fluorine for crystal growth, via prefluorination
of oxide compounds with hydroxyl acid.
YLF single crystals doped with Yb ions at three different doping
levels (5 at%, 7.5 at%, and 10 at%) were grown. For each growth,
the mixture of starting materials was prepared inside a moisture-
controlled environment, storing the components, in the proper
relative concentrations, inside a dedicated high-purity platinum
crucible (99.9%). The relative concentrations of host components
used were 52.5 mol% LiF and 47.5 mol% YF3. A suitable concentration
of YF3 was replaced by YbF3 in order to achieve the desired doping
level (5 at%, 7.5 at%, and 10 at%). Total amounts of raw materials
between 80 g and 100 g were used for all the growths.
Prior to load the starting materials inside the growth chamber,
the furnace was carefully precleaned. Suitable solvents were firstly
used. Subsequently, the furnace was evacuated at low pressure
and then heated up to 1000°C for several hours while drawing
with the vacuum pumps. An ultimate vacuum level of 10–6 Pa was
reached. After high-vacuum cleaning, the crucible, containing the
mixture of starting materials, was loaded inside the furnace. The
growth atmosphere was built up pumping inside the furnace argon
and CF4 gases in the proper concentrations. The furnace was then
heated up until the melting point of the starting materials was
reached. The melt temperature slightly changed as a function of
the Yb doping level, between 880°C and 890°C. Impurities emerged
on the melt surface were removed by using high-purity (99.9%) Pt
tubes or crystal seeds, repeatedly plunged into the melt at different
temperatures. When the melt surface appeared completely free from
foreign particles, the growth was started by lowering a crystal seed
in contact with the melt. The crystal seed was slowly lowered toward
the melt surface in order to prevent thermal shock and sudden
uncontrollable crystallizations of the surface. For all the growths,
a YLF undoped seed, oriented along the a axis, was used (see Sec.
11.5.2.1). The molten was slightly undercooled by a few Celsius
degrees, below the melting point, when the nucleation started. In the
early stage of the growth, fine temperature changes were operated
to shrink the crystal diameter down to 1–2 mm. After the growth
of a narrow neck, a few millimeters long, the crystal diameter was

smoothly enlarged up to 1–2 cm. Subsequently, the growth was
carried at a uniform diameter for 5–10 cm. For all the samples the
pulling rate was 0.5 mm/h, leading to growth times of several days
for each boule. The rotation rate was kept at 5 rpm.
In Fig. 11.9 are shown typical temperature and diameter data
acquired during the growth process. Such graphs show typical
temperature changes and diameter variation rates for a good
growth process. As can be seen from graphs, the melt temperature
changed within a few Celsius degrees during the growth process
and diameter fluctuations achieved in growth conditions with
the PID control are of the order of 1%. The origin for both graphs
corresponds to the nucleation. The temporal evolution graph for
crystal diameter shows the diameter decrease, operated in the early
stage of the growth, for the necking operation, by changing the melt
temperature. After the growth of a narrow neck for a few millimeters,
the diameter of the growing crystal was enlarged at very slow rates
by using the PID control. The red line in the graph represents the set
point of the optical diameter control. Blue data points correspond
to the experimental measurements of the diameter. A very slow rate
was applied to enlarge the diameter of the growing crystal in order
to prevent formation of defects due to irregular anisotropic growth
rates. The largest diameter value was set around 1 cm. The growth at
uniform diameter was carried on for several centimeters. Along with
temperature and diameter graphs, in Fig. 11.9 is shown a picture of
the finished crystal.
After growth, before being extracted from the furnace, the crystal
boules were slowly cooled down to room temperature. Typical
cooling rates used to prevent thermal shock were 15°C/h.
As soon as finished crystals are removed from the furnace,
some defects can be visually detected. Good-quality Yb-doped
fluorides appear transparent, as the Yb3+ ions do not have any
visible absorption bands and the host is transparent from the UV
(180 nm) up to several microns. The regular growth along the a
axis results in elliptic cross sections with two well-defined lateral
edges. Irregular diameter changes are typical indications of internal
defects. Polycrystalline domains appear unclear and cracked. In Fig.
11.10 are reported some pictures of grown boules of YLF:5%Yb,
YLF:7.5%Yb, and YLF:10%Yb. All the crystal boules appear clear and
transparent. In Fig. 11.10b it can be seen the crystal seed with the Pt
wire, used to fasten the crystal seed to the seed holder extension, still
attached to the crystal boule. In both Fig. 11.10c and Fig. 11.10d is
visible the narrow crystal neck operated in the early stage of growth
process by changing the melt temperature. Narrow diameters of 1–3
mm (Fig. 11.10c) and 2–3 mm (Fig. 11.10d) were grown for up to
10 mm (Fig. 11.10c). The lateral crystals edges and the elliptic cross
section can be observed in Fig. 11.10a,c,d. In Fig. 11.10e the circular
cross section is indication of a c axis component along the growth
axis. Irregular diameter changes as those visible in Fig. 11.10a,b,d
did not reflect in the inclusion of defects. An example of excellent
diameter stability is provided in Fig. 11.10c.
867.5
867.0
866.5
Temperature [°C]
866.0
865.5
865.0
864.5
20 40 60 80 100 120 140 160 180
Time [h]
14
12
10
Diameter [mm]
0 20 40 60 80
Length [mm]
Figure 11.9 Temperature and diameter data relative to the growth of a

YLF:5%Yb single crystal along with a picture of the finished
boule.
(b) (c)
(a)
(d) (e)
Figure 11.10 As-grown crystals of YLF:5%Yb (a and b), YLF:7.5% (d), and
YLF:10%Yb (c and e).
In Fig. 11.11 is highlighted the effect of a faulty growth atmos-

phere composition on the optical quality of Yb-doped single crystals.
In Fig. 11.11a is shown the formation of Yb2+ impurities inside Yb-
doped fluorides due to a CF4-depleted growth atmosphere. Crystals
grown under this kind of atmosphere incorporate Yb2+ ions, which
are responsible of the red coloration and hence of the poor optical
properties. Such samples are completely useless for optical cooling
experiments. On the other side, in Fig. 11.11b is shown a carbon
scum deposition both on the crystal surface and on the interior walls
of the furnace, due to an excess of CF4 concentration in the growth
atmosphere. Such growth conditions are rather detrimental for the
growth process as carbon impurities can be incorporated inside the
growing crystals. Moreover, the carbon scum which deposits on the
interior of the furnace typically accumulates also on the surfaces of
the top windows, making difficult the optical control of the growing
crystal diameter. Irregular diameter variations typically occur.
(a) (b)
Figure 11.11 (a) Yb2+ contamination of Yb-doped YLF single crystals grown
under a CF4-depleted atmosphere. (b) Carbon deposition
on crystal surface and furnace interior due to excessive
concentration of CF4 in the growth atmosphere.
11.5.2.3 Preparation of crystal seeds for the growth of Yb-

doped YLF single crystals
Prior to the growth of the Yb-doped samples, an undoped YLF
single crystal was grown in order to fabricate crystal seeds for the
growth of the cooling materials. For the growth of the undoped YLF,
the mixture of starting materials was composed of LiF and YF3 raw
powders mixed with the relative concentration 52.5 mol% LiF and
47.5 mol% YF3. A high-purity (99.9%) platinum crucible dedicated
to undoped materials was used. The growth atmosphere was
composed 100% of high-purity (5N) argon. Crystal seeds obtained
from previous growths of undoped YLF single crystals were not used
in order to prevent any inclusion of impurities, not contained in the
LiF and YF3 raw powders. The growth was hence started, inducing
self-nucleation around a high-purity (99.99%) Pt wire wounded up
in a tubular shape with a narrow diameter. A slightly higher melt
temperature are typically necessary to induce the crystallization
starting from a Pt wire instead of a crystal seed. During the growth,
the crystal was pulled at a rate of 1 mm/h while rotated at a speed
of 5 rpm. A picture of the finished boule is shown in Fig. 11.12a. The
circular cross section of the crystal boule is indication of growth
axis mostly aligned along the c axis of the unit cell. In Fig. 11.12b
are shown some crystal seeds fabricated from the grown crystal
boule. Such samples are typically used in parallelopiped shapes
with edges oriented along the optical axes and tip-cut ends. Typical
dimensions are 3 × 3 (c × a) mm2 cross section and 5–10 mm (a)
length. The longer axis of the parallelopiped was cut parallel to the
a axis. This configuration defines a crystal seed oriented along the
a axis. The crystal seed is typically mounted on the seed holder in
order to contact the melt with the longer axis perpendicular to the
melt surface. A crystal seed oriented along the a axis hence induce
oriented nucleation with the growth axis of the growing crystal
aligned with the a axis of the unit cell.
Although the self-nucleation preferentially tends to occur along
the c axis, the choice of oriented growth along the a axis is related to
the optical quality of finished crystals.
The precise orientation of the crystallographic axes of crystal
boule were previously determined by X-ray diffraction and the
absence of major impurities and scattering centers was checked
through spectroscopic and scattering analysis, as detailed in Section

11.3.
(a) (b)
Figure 11.12 (a) As-grown crystal boule of YLF undoped crystal. (b) Seeds of
YLF undoped crystal oriented along the a axis.
11.5.2.4 The growth of Yb-doped YLF single crystals with

controlled Tm doping
A novel scheme for laser cooling, relying on virtuous energy
transfer processes, has been investigated in Yb–Tm co-doped
YLF single crystals, providing the first experimental evidence
of energy transfer–assisted anti-Stokes cooling [21]. Significant
efficiency enhancements, over the single Yb doping performances,
were achieved via Yb–Tm energy transfer in the co-doped single
crystal. The virtuous energy transfer process between Yb and Tm
ions provides an additional channel of heat removal via radiative
emission at higher average energies than absorbed photons. The
potentialities of this approach for intrinsic efficiency enhancements
are huge.
A YLF single crystal Yb doped at 5 at% with a controlled Tm
doping of 16 ppm was grown. A low concentration of Tm ions was
chosen as the starting point for the investigation of the effect of
Yb–Tm energy transfer on the Yb anti-Stokes process.
The co-doped sample was grown under similar conditions as
those used for the single Yb-doped samples. For the mixture of
starting material, the usual concentrations of host components were
used: 52.5 mol% LiF and 47.5 mol% YF3. However, to achieve the
Yb–Tm co-doping, the YF3 compound was partially replaced with
appropriate concentrations of both YbF3 and TmF3 compounds. The
relative concentrations were calculated in order to replace 5% of Y3+
host ions with Yb3+ ions and 0.0016% of Y3+ host ions with Tm3+ ions.
Due to the similarity of Yb3+ and Tm3+ ionic radii sizes, a segregation
coefficient equal to 1 is expected, even for the substitution of Y3+/
Tm3+ substitution [11].
The growth process was carried out under an argon and CF4
atmosphere. During the growth, the pulling rate was 0.5 mm/h and
the rotation speed 5 rpm. As a crystal seed, a YLF undoped single
crystal oriented along the a axis was used.
Due to the low concentrations of the TmF3 compound required for
the desired doping level, the actual concentration of Tm was checked
after the growth with a trace element analysis performed by means
of laser ablation–inductively coupled plasma–mass spectrometry
(LA-IPC-MS) [20]. A Tm concentration of 16 ppm was detected. No
internal defects related to the Yb–Tm co-doping were detected.
11.5.3 Investigation of Sample Quality: Structural

Analysis, Testing of Internal Defects, and Major
Contaminants
After growth, the crystal boules were investigated with different
techniques in order to assess the optical and structural quality. The
local single-crystalline structure and the orientation of the unit cell
were investigated through X-ray backscattering diffraction in a Laue
chamber. The inclusion of structural defects was assessed observing
the propagation of visible laser beams through the volume of
samples. Absorption measurements were performed to check the
absence of major contaminants.
11.5.3.1 Structural analysis: X-ray backscattering diffraction

Due to the optical anisotropy of the YLF crystal, its spectroscopic
properties depend on the crystal orientation and hence need to be
selectively investigated for all the independent polarizations. The
orientation of the crystallographic axes, and hence of the optical
axes, needs to be accurately identified so that samples oriented
along the optical axes can be fabricated maximizing absorption and
emission cross sections of the materials.
After growth, all the crystal boules were firstly analyzed by X-ray
backscattering diffraction in a Laue chamber. Such analysis enables
to identify the orientation of the crystallographic axes of the unit
cell with high accuracy. Moreover, diffraction patterns acquired in

a different region of the samples allow the checking of the single-
crystalline structure of the sample.
For the Laue analysis, the sample, fixed on a goniometer holder,
is shined with a nonmonochromatic X-ray beam. The precision
goniometer, which allows rotations around the three axes, is
used to fix the orientation of the crystal with respect of the X-ray
direction. The backscatter radiation is collected on a sensor plate,
placed between the X-ray source and the crystal, which record the
diffraction pattern.
After acquisition of the diffraction pattern in the Laue chamber,
suitable software is used to simulate the backscattering of the crystal
for a given orientation of the crystallographic axes with respect to the
direction of the X-ray beam. Combined rotations around the three
axes are applied in order to reproduce the experimental pattern. The
correspondence between the acquired pattern and the simulation
allows one to identify the orientation of the crystallographic axes,
with an accuracy of 1°.
The absence of a polycrystalline domain can be locally checked
from the diffraction pattern. The presence of large or double spots
instead of single dots in the backscatter pattern is due to diffraction
from a staggered lattice and hence indicates a polycrystalline
structure.
Some examples of a diffraction pattern relative to the grown
crystals are shown in the following section.
In Fig. 11.13a is shown the diffraction pattern acquired in
backreflection for the YLF:7.5%Yb boule shown in Fig. 11.10d, by
shining the X-ray beam parallel the direction of the two lateral edges.
Along with the measured diffraction pattern is shown the theoretical
pattern, simulated for the c axis parallel to the X-ray direction. The
correspondence between the two patterns confirms the growth axis
aligned with the crystallographic a axis and returned the orientation
of the crystallographic axes with accuracy of 1°. Single dots in the
diffraction patterns acquired in different areas of the boule ascertain
the single-crystalline structure.
In Fig. 11.13b is shown the diffraction pattern acquired for the
YLF:10%Yb crystal shown in Fig. 11.10e, by shining the X-ray beam
in proximity of the single edge. The X-ray simulation returned a c
axis component along the growth axis, as suggested by the circular
cross section of the sample. The growth axis resulted in the plane c-a,
tilted about 30° with respect to the a axis. In Fig. 11.14 is reported
the diffraction pattern acquired for the as-grown crystal boule of YLF
undoped, by shining the X-ray beam perpendicularly to the growth
axis.
a
a c
a
c a (a) c a (b)
Figure 11.13 (a) X-ray backscatter diffraction pattern of YLF:7.5%Yb

(sample shown in Fig. 11.10d), acquired by shining the X-ray
beam along the direction of the two lateral edges, and relative
simulation. (b) Experimental pattern, acquired by shining
the X-ray beam in proximity of the lateral edge, and relative
simulation for the YLF:10%Yb sample shown in Fig. 11.10e.
a c
Figure 11.14 Acquired backscatter diffraction pattern and relative

simulation of as-grown YLF undoped crystal. The X-ray beam
was shone perpendicularly to the growth axis.
The correspondence with the theoretical pattern simulated

for the c axis along the growth axis indicated that the favorable
growth axis for the self-nucleation in YLF crystal is the c axis. The
measurement of the crystallographic axes orientation returned the c
axis tilted just a few degrees with respect to the growth axis.
11.5.3.2 Scattering analysis for crystal defect detection

Internal defects, such as cracks, microbubbles, and dislocations
can be detected observing the propagation of a laser beam through
the volume of the sample [1]. Crystal defects give rise to scattering
centers, hence scattering phenomena of laser beams are indication
of internal defects. In the case of good crystal quality, scattering
is absent. Laser wavelengths not absorbed by the material under
investigation should be chosen. For Yb-doped fluorides all the visible
wavelengths are suitable for the scattering analysis.
A very sensitive test for internal strains in uniaxial crystals can
be performed observing the extinction ratio of the sample, placed
between two crossed polarizers with the c axis along the direction
of propagation of light. In the case of strain, a significant fall of the
extinction ratio is detectable.
As-grown boules and finished samples were tested, observing the
propagation of visible laser beams by means of a Zeiss microscope.
Two different wavelengths were used: 633 nm and 405 nm. In Fig.
11.15 are shown some tests of crystal quality performed with the
633 nm laser beam.
In Fig. 11.15a and Fig. 11.15b is shown the propagation of the
633 nm laser beam inside a cooling sample that possesses all the
surfaces polished at a high optical grade. In Fig. 11.15a a single
scattering center is detected, and reflections at the entrance and exit
surfaces are visible. In Fig. 11.15b is shown a defect-free area of the
same sample. Figure 11.15c shows a dense aggregate of scattering
center detected in a crystalline boule grown under an unsuitable
thermal gradient. On the bottom of Fig. 11.15d–f is shown the axial
distribution of scattering center detected in a defected crystalline
boule. In Fig. 11.15d is shown the propagation of the laser beam
through the neck: a high density of scattering center is visible. In Fig.
11.15e is shown the laser propagation in the middle of the boule:
scattering defects are less numerous in this region.
In Fig. 11.15e is shown the bottom of the boule: scattering
is almost absent. The 405 nm wavelength was used to check the
absence of scattering center in finished cooling samples, where the
polished surfaces strongly reduce spurious reflections. Moreover,

the shorter laser wavelength increases the scattering cross section,
enabling a more refined analysis of the propagation of the laser
beam.
(a) (b) (c)
(d) (e) (f )
Figure 11.15 Test of crystal quality with 633 laser beam. (a) Single scattering
center detected inside a cooling sample; (b) scattering-
free area of the same sample; (c) high-density aggregate of
scattering centers; (d) propagation of 633 nm laser beam
in the neck of a crystalline boule: high density of scattering
defects; (e) middle region: less numerous scattering center;
and (f) bottom: scattering almost absent.
Internal stress in as-grown crystal boules were investigated by

means of crossed polarizers. No significant variation of the extinction
ratio were detected, indicating good crystal growth conditions.
11.5.3.3 UV-NIR absorption measurement for detection of

major contaminants
Inclusion of major contaminants inside grown crystals can be
examinated through absorption measurements. Spurious bands
in the absorption spectrum of the sample are the fingerprints of
optically active impurities.
Among rare earth ions, the Yb3+ ion possesses a peculiar energy
level structure, which exhibits a single excited manifold 2F5/2 at
energies between 1.2 and 1.3 eV with respect to the ground-state
2F
7/2. In the absence of major contaminants only the absorption
band relative to the Yb transition, which extends from 890 nm to
1100 nm in fluorides, should be detectable. Any additional line in the

absorption spectrum can be ascribed to undesired impurities.
As detailed earlier, the most difficult issue when growing Yb-
doped fluorides is to avoid the presence of oxygen complexes and
Yb2+ ions in the melt. When hydroxyl radicals, OH–, substitute
F– ions in fluoride crystals they give rise to UV electronic bands
between 180 and 230 nm and vibrational absorption in the range
of 3000–4000 cm–1 [3]. The inclusion of OH– impurities in fluoride
crystals, however, is typically detectable from the analysis of the UV
absorption spectrum as in this region OH– complexes exhibit the
strongest absorption. The vibrational absorption in the IR is much
weaker than UV electronic bands and typically can be detected only
at low temperatures. Low-temperature measurements, however,
normally are performed when UV absorption bands are detectable
in order to provide further support of OH– contaminations. Yb2+ ions
in fluoride crystals instead possess large absorption bands between
200 and 600 nm due to 4f–5d transitions and hence can be detected
from the absorption spectrum in this range.
In addition to oxygen complexes and Yb2+ ions, impurities of
other rare earth and transition metals can be also incorporated
inside growing crystals due to small inaccuracies in the cleaning
procedures.
To check the absence of major contaminants, for all the grown
samples, a broad absorption spectrum from the UV (180 nm) to the
NIR (2000 nm) was acquired. Figure 11.16 reports the absorption
spectra acquired for the Yb-doped YLF samples. All the spectra
exhibit a single absorption band identifiable with the Yb 2F7/2 Æ
2F
5/2 transition.
No additional bands can be detected, indicating that the
inclusion of major impurities can be ruled out. In particular, the UV
spectrum indicates the absence of OH– contaminants in relevant
concentrations, that is, higher than tens of ppm. The absence of a
spurious band in the whole visible range proves the efficiency of
the growth atmosphere in preventing formation of Yb2+ ions in
the melt. The inclusion of other kind of contaminants in relevant
concentrations can be ruled out as well.
The absorption spectrum of the Yb–Tm co-doped sample showed
a similar behavior as that of the single Yb-doped YLF single crystals.
Figure 11.16 Absorption spectra ranging from 180 nm up to 2 µm of

YLF:5%Yb, YLF:7.5%Yb, and YLF:105Yb samples. The
wavelength resolution is 0.4 nm between 180 and 800 nm
and 0.6 nm between 800 and 2000 nm, while the absorption
resolution is 0.002 cm–1.
The multipeak structure of the Yb3+ absorption band is related to

the interaction of Yb3+ ions with the electric field of the surrounding
crystal, which induces a Stark splitting of the free ions’ energy levels
of 102 cm–1 magnitude. For the Yb3+ ions occupying the S4 sites in the
YLF host the excited manifold 2F5/2 is divided in three sublevels and
the ground-state 2F7/2 in four sublevels. Phonon-assisted inter-Stark
transitions between these sublevels are the origin of the multiple
lines visible in the Yb absorption band at room temperature.
The increase of the background absorption between 300 nm and
180 nm is not due to the host absorption, which is transparent in
this region, but mainly to Rayleigh scattering from the surfaces. Even
if the entrance and exit facets of the samples were polished before
absorption measurements, surface roughness and imperfections
are still present and cause efficient scattering at short wavelength,
increasing the background absorption in such region. The automatic
detector change around 800 nm is the source of the noise in this
region.
11.5.4 Sample Fabrication

After identification of the crystallographic axes orientation and
assessment of the absence of polycrystalline domains, internal
defects, and major contaminants, samples for cooling experiments
were prepared. Precision cutter machines, equipped with a diamond
wire saw, were used to cut samples of the desired orientation and
dimensions from the crystals boules. Right angle–cut and Brewster-
cut samples were fabricated.
Right angle samples were cut to a parallelopiped shape, with
the longer edge parallel to the a axis in order to allow the excitation
along the c axis, which maximizes the absorption in YLF crystals. A
layout is shown in Fig. 11.17a. Typical dimensions were 3 × 3 mm2 (c
× a) cross section and length (a) between 5 and 15 mm.
Brewster-cut cooling samples were fabricated to minimize
reflection losses. The orientation was designed to eliminate
reflections for incident light polarized along the c axis of the crystal.
Pictures of finished right angle–cut and Brewster-cut samples are
shown in Fig. 11.17 and Fig. 11.18.
(a) (b)
Figure 11.17 (a) Layout of right angle–cut sample prepared for cooling
experiments in single-pass configuration. (b) Picture of
finished cooling sample of YLF:10%Yb.
All the surfaces of cooling samples were polished to a high

optical grade using fine diamond powders. The quality of the crystal
surfaces significantly affects the fluorescence escape efficiency;
hence high–optical grade polishing is of fundamental importance.
Figure 11.18 Picture of finished Brewster-cut sample of YLF:5%Yb for laser

cooling experiments.

Several Yb-doped fluoride single crystals have been grown in our
facility for laser cooling experiments. Growth parameters have
been optimized for this kind of materials achieving high optical
and structural quality. The remarkable performances achieved in
bulk cooling experiments confirmed the high crystal quality. Crystal
growth techniques and methods for optical and structural quality
assessments were reviewed.
Acknowledgments
The authors would like to acknowledge I. Grassini for her competence
and care in preparing the samples and D. Parisi and S. Veronesi for
useful discussions and their expertise in crystal growth. A. Volpi
acknowledges support by the European Space Agency under grant
no. 4000108074/13/NL/PA, “Cooling Effect on Fluoride Crystals.”
References 375
References
1. Belt, R. F., Uhrin, R. (1991). Top seeded solution growth of Cr3+:LiCaAlF6

in HF atmosphere, J. Cryst. Growth, 109, pp. 334–339.
2. Brice, J. C. (1973). The Growth of Crystals from Liquids (Series of
Monographs on Selected Topics in Solid State Physics) (North Holland,
Amsterdam).
3. Cornacchia, F., Palatella, L., Toncelli, A., Tonelli, M., Baraldi, A.,
Capelletti, R., Cavalli, E., Shimamura, K., Fukuda, T. (2002). Temperature
dependence of impurity quenched luminescence in Tm3+:LiLuF4, J.
Phys. Chem. Solids, 63, pp. 197–202.
4. Czochralski, J. (1918). A new method for the measurement of the
crystallization rate of metals, Z. Phys. Chem., 92, pp. 219–221.
5. Di Lieto, A., Sottile, A., Volpi, A., Zhang, Z., Seletskiy, D. V., Tonelli, M.
(2014). Influence of other rare earth ions on the cooling efficiency in
Yb:YLF crystals, Opt. Express, 22, pp. 28572–28583.
Stokes scattering, Phys. Rev. Lett., 46, pp. 236–239.
7. Epstein, R. I., Buchwald, M., Edwards, B., Gosnell, T., Mungan, C. (1995).
Observation of laser-induced fluorescent cooling of a solid, Nature,
377, pp. 500–502.
8. Fisher, S., Ivaturi, A., Frolich, B, Rudiger, M., Richter, Kramer, K. W.,
Richards, B. S., Goldschmidt, J. C. (2014). Upconverter silicon solar
cell devices for efficient utilization of sub band gap photons under
concentrated solar radiation, IEEE J. Photovolt., 4, pp. 183–189.
9. Fisher, S., Favilla, E., Tonelli, M., Goldschmidt, J. C. (2015). Record
efficient upconverter solar cell devices with optimized bifacial silicon
solar cells and monocrystalline BaY2F8:30%Er upconverter, Sol.
Energ. Mater. Sol. Cells, 136, pp. 127–136.
10. Hehlen, M. P., Boncher, W. L., Melgaard, S. D., Blair, M. W., Jackson, R. A.,
Littleford, T. E., Love, S. P. (2014). Preparation of high-purity LiF, YF3
and YbF3 for laser refrigeration, Proc. SPIE, 9000, p. 900004.
11. Kaminski, A. A. (1996). Crystalline Lasers: Physical Processes and
Operating Schemes (CRC Press, Boca Raton, FL).
12. Melgaard, S. D., Seletskiy, D. V., Di Lieto, A., Tonelli, M., Sheik-Bahae,
1588–1590.
13. Melgaard, S. D. (2013). Cryogenic Optical Refrigeration: Laser Cooling

of Solids Below 123K, PhD dissertation, University of New Mexico,
Albuquerque, NM.
14. Melgaard, S. D., Albrecht, A., Hehlen, M. P., Seletskiy, D. V., Sheik-Bahae,
M. (2014). Optical refrigeration cools below 100 K CLEO 2014, OSA
Technical Digest, paper FTh4D.4.
15. Pamplin, B. (1975). Crystal Growth (Pergamon Press, Oxford, UK).
Lumineszenz-und Temperaturstrahlung, Z. Phys., 57, pp. 739–746.
17. Seletskiy, D. V., Melgaard, S. D., Bigotta, S., Di Lieto, A., Tonelli, M., Sheik
18. Thiede, J., Distel J., Greenfield, S. R., Epstein, R. I. (2005). Cooling to 208
K by optical refrigeration, Appl. Phys. Lett., 86, p. 154107.
19. Thoma, R. E., Weaver C. F., Friedman, H. A., Insley, H., Harris, L. A., Yakel,
H. A. (1961). Phase equilibria in the system LiF-YF3, J. Phys. Chem., 65,
pp. 1096–1099.
20. Tiepolo, B., Bottazzi, P., Palenzona, M., Vannucci, R. (2003). A laser probe
coupled with IPC - double focusing sector: field mass spectrometer
in situ analysis of geological samples and U-Pb dating of zircon. Can.
Mineral, 41, pp. 259–272.
21. Volpi, A., Di Lieto, A., Tonelli, M. (2015). Novel approach for solid state
cryocoolers, Opt. Express, 23, pp. 8216–8226.
Chapter 12
Microscopic Theory of Optical

Refrigeration of Semiconductors
Rolf Binder and Nai-Hang Kwong

College of Optical Sciences, University of Arizona, Tucson, AZ 85721, USA
binder@optics.arizona.edu
This chapter is intended to give an overview of some theoretical

efforts to analyze optical refrigeration of inorganic semiconductor
crystals. A fermionic theory describing the conduction and valence
bands, combined with Coulomb interaction between the excited
charge carriers, yields the absorption and luminescence spectra
that include excitonic effects valid for arbitrary excitonic ionization
ratios. The microscopic theory used as an input to the Sheik-Bahae
and Epstein semiconductor cooling model yields a cooling analysis
that has been evaluated for GaAs. Extensions of the basic cooling
model for bulk crystals include light propagation and luminescence
re-absorption effects and the effects of passivation layers in doped
spatially inhomogeneous structures. While the numerical results
presented here are restricted to GaAs, the general theoretical
formulation is valid for any direct-gap crystalline semiconductor.

www.panstanford.com
378 Microscopic Theory of Optical Refrigeration of Semiconductors
12.1 Introduction
The conceptual foundation of optical refrigeration of solids was laid
by Pringsheim [1]. Soon after the first major experimental break-
through by Epstein et al. [2], who achieved net cooling of an Yb-
doped ZBLANP glass, the study of the physics and applications of
optical refrigeration of solids became a major worldwide research
area, as documented by review articles, including those written by
Sheik-Bahae and Epstein [3] and Nemova and Kashyap [4]. Works
following the 1995 break-through also included research on the
cooling of inorganic semiconductors, which are generally believed
to hold promise for cooling to temperatures of about 10 K. The
semiconductor research includes various theoretical aspects, for
example those discussed by Oraevsky [5], Rivlin and Zadernovsky
[6], Sheik-Bahae and Epstein [7], Apostolova [8], and Khurgin [9],
as well as experimental studies, including, for example, the ones
presented by Gauck et al. [10] and Imangholi et al. [11]. Important
recent milestones are the observation of cooling to cryogenic
temperatures of Yb-doped YLF crystals reported by Seletskiy et al.
[12] and the cooling of nanostructured CdS crystals reported by
Zhang et al. [13].
Inorganic semiconductors, such as GaAs or CdS crystals, are
commonly characterized by their electronic band structure. In the
vicinity of the fundamental bandgap, the band structure can be
approximated by two bands, the conduction and the valence band
(Fig. 12.1). However, optical transitions involved in absorption and
photoluminescence processes are generally not well described by
band-to-band transitions, owing to the strong Coulomb interaction
between electrons in the conduction band and holes in the valence
bands. The Coulomb interaction can give rise to sharp optical
resonances called excitonic resonances, which correspond to
the bound electron–hole pairs (excitons). Optical refrigeration
of semiconductors is dominated by excitonic resonances, both
in the absorption (pumping) and the upconverted (anti-Stokes)
luminescence process. While at low excitation densities (low
pumping intensities) the excitons can be described with simple
Bosonic models, such descriptions are not appropriate for optical
refrigeration, as the strong pumping gives rise to relatively large
excitation densities and hence to the partial ionization of excitons.
Theoretical Foundation and the Importance of Excitonic Effects 379
A comprehensive and predictive theory must therefore be built

upon a Fermionic description of electrons and holes and must
be nonperturbative in the Coulomb interaction to allow for the
modeling of excitons or, more precisely, a partially ionized exciton
gas. We have developed such a microscopic theory and applied to
the analysis of several questions important for the understanding of
realistic experimental setups.
In this chapter, we will first review the microscopic theory and
its basic results, which is the dominant role of excitons played in
the cooling at low (few tens of Kelvin) temperatures. We then apply
the theory to the description of realistic structures in which the
propagation of light (in particular luminescence light) gives rise to
effects such as luminescence re-absorption and finite luminescence
extraction from the system. Furthermore, in many experiments
the pump beam does not cover the entire cross section of the
semiconductor crystal, and we study effects of finite lateral beam
size on the cooling. Finally, we study more complex systems in which
the semiconductor crystal is doped and which additionally may be
sandwiched between doped passivation layers, which are intended
to reduce nonradiative surface recombination.
12.2 Theoretical Foundation and the

Importance of Excitonic Effects
We begin by summarizing our theoretical model for the optical
absorption and photoluminescence spectra and their use in the
cooling model of Sheik-Bahae and Epstein [7]. Details on the
derivation of our theory can be found in Kwong et al. [14], as well as
Rupper et al. [15, 16]. As mentioned above, an important question
that our theory can answer is this: Given the fact that excitonic
resonances dominate semiconductor absorption and luminescence
spectra (especially at low temperatures), is the same true for the
cooling characteristics? In other words, do we need excitonic effects
to achieve cooling at low temperatures?
Starting from an electronic band structure model in which we
have (possibly degenerate) valence (v) and conduction (c) bands,
which are parabolic and isotropic, we use a microscopic Green’s
function theory that is nonperturbative in the Coulomb interaction
between the charge carriers (electrons in the conduction band and

holes in the valence band), most importantly between the electrons
and holes, and hence carries full information about the excitonic
resonances (bound states of electrons and holes) without treating
the exciton as a separate quasi-particle (as is often done in Bosonic
models for the excitons).
In this section we start out by analyzing a spatially homogeneous
semiconductor structure, where in the presence of a partially
ionized exciton gas the frequency-dependent absorption coefficient
α(ω) is given in terms the retarded interband susceptibility χR(q, ω),
evaluated in the long-wavelength approximation at approximately
zero wavevector q,
a (w ) =
4pw
n bc
Im (c R
) (12.1)
(q = 0, w )
where nb is the background dielectric index and c the speed of light

in vacuum. Once the absorption coefficient is known, the Kubo–
Martin–Schwinger (KMS) relation, Zimmermann [17], can be used
to relate absorption to luminescence. In the present case of a bulk
semiconductor the the KMS relation reads
2
Ê wn ˆ
R(w ) = Á b ˜ g(w )a(w ) (12.2)
Ë pc ¯
This relation is valid if the system is in quasi-thermal equilibrium
at temperature T. Here R(ω) is the luminescence spectral density,
g(ω) is the Bose function,
1
g(w ) = (12.3)
( w - me - mh )/kBT
e -1
where μa is the chemical potentials of electrons (a = e) and holes (a
= h), respectively, and μB is the Boltzmann constant. If, instead of
a bulk semiconductor, we have a semiconductor quantum well, the
KMS relation reads [18]
2
2 Ê wn ˆ
R(w ) = Á b ˜ g(w )Ims R (w ) (12.4)
3 Ë pc ¯
where the retarded response function ImσR(ω) is approximately
given by the absorbance AOW(ω) of the quantum wells, which is
defined in terms of reflectance and transmittance as
Ir I t
AQW (w ) ∫ 1 - - ª Ims R (w ) (12.5)
I0 I0
where I0, Ir, and It are the incident, reflected, and transmitted
intensities, respectively.
ε
c
k
v
Figure 12.1 Sketch of parabolic band structure (energy versus wavevector)

for valence (v) and conduction (c) bands.
The retarded interband susceptibility entering Eq. 12.1 follows

from an integral equation that, in our approximation, corresponds
to the so-called T matrix approximation, which is a nonperturbative
infinite summation of ladder diagrams. It reads
c R (q ,w ) = c 0R (q ,w )
- Â
sj (k ¢ , q , w )d js (k ¢ )
d js (k )gsj0R (k , q ,w )TsjR (k , k ¢ , q ,w )g0R
sjk k '
(12.6)
where the summation is of the band indices labeling the spin-
degenerate conduction bands (index s) and spin-degenerate valence
bands (index j), and over the wavevectors k and k¢. The susceptibility
Eq. 12.6 describes the propagation of electron–hole pairs excited
by photon absorption. In the “free” part χ 0R(q, ω), the electron and
hole propagate independently of each other, and the T matrix TsjR
represents interactions between the two. Explicitly,
c 0R (q ,w ) = - Âd
js (k ) gsj (k , q , w )d js (k ) (12.7)
0R
sjk
where
djs(k) (12.8)
denotes the interband dipole matrix element (for details of this and
related matrix elements see Gu et al. [19]). The imaginary and real
parts of the free two-particle Green’s function are given by
{ }
sj (k , q , w ) = -
Im g0R
1 dw ¢
2 2p Ú { }
1 - f s (w - w ¢ ) - f j (w ¢ )
(12.9)
¥ As (k s ,w - w ¢ )A j (k j ,w ¢ )
and
{ } Ú { }
dw ¢ Im gsj (k , q ,w ¢ )
0R
(12.10)
sj (k , q , w ) = P
Re g0R
p w¢ - w
where
1 (12.11)
fa (w ) = ( w - ma )/kBT
e +1
are the Fermi functions for electrons (a = e) and holes (a = h), and the
spectral functions are given by
{
Aa (k ,w ) = -2Im GaR (k ,w ) (12.12) }
In its present stage, our theory does not include a microscopic
description of phonons and electron–phonon coupling. However,
for optical refrigeration it is important to account for excitonic line-
broadening effects since the optical pump beam is usually chosen
to be in resonance with the low-frequency tail of the lowest exciton
resonance. We therefore include a phenomenological amendment to
the spectral function that accounts for such line-broadening effects.
The temperature dependence of the line broadening due to coupling
of electrons to longitudinal acoustic (LA) and longitudinal optic (LO)
phonons is approximately given by
g LO
g ph
0
(T ) = g LAT + WLO / kBT
(12.13)
e -1
This yields the following phonon-broadened spectral functions
for electrons and holes:
dw ¢
Ú 2p
2
Aaph (k ,w ) = Aa (k ,w - w ¢ ) (12.14)
pg ph
0
cos h(2w ¢/g ph
0
)
The retarded (superscript R) single-particle Green’s function is
given by

GaR (k ,w ) =
w - e a (k ) - S aR,T (k ,w ) - S R,exc
a (k ) - S R,CH
a (k )
(12.15)
where the unrenormalized single-particle energies (i.e., the
parabolic band structure, parametrized in terms of effective masses
ma) is given by
2
2 k a
e a (k a ) = (12.16)
2ma
and the real and imaginary parts of the T matrix (superscript T) self-
energies (see Kwong et al. [14] and also Schmielau et al. [20] and
Piermarocchi and Tassone [21]) are
Re {
S Ra ,T (k ,w )} Ú =P
R ,T
{
dw ' Im S a (k ,w ¢ )
(12.17)
}
p w¢ - w
and
{
Im S R,T
a (k , w ) = } Â Ú d2wp¢ {g(w + w ¢) + f (w ¢)}

a¢
a¢ k ¢ (12.18)
¥ Im {T ¢¢ (p , p , q ,w + w ¢ )} A (k ¢ ,w ¢ )
R
aa¢ a¢
where q = k + k¢ and
ma¢ k - ma k ¢
p= (12.19)
ma + ma¢
Furthermore, the real-valued exchange (superscript “exc”) self-
energy is given by
dw
S R,exc
a (k ) = - ÂW
k¢
aa (k - k¢ )Ú 2p f (w )A (k ,w ) (12.20)
a a
and the density of the charge carriers follows from

dw
na = Â Ú 2p f (w )A (k ,w ) (12.21)
k
a a
which is an implicit equation for the chemical potential μa. The

Coulomb hole (superscript CH) self-energy is
S R,CH
a =- Â (W
k
aa (k ) - V (k ) ) (12.22)
The “ladder” equation for the T matrix reads

Taa ¢ (k , k ¢ , q , w ) = Waa ¢ (k - k ¢ ) +
R

ÂW
k ¢¢
aa ¢ (k - k ¢¢ , w ) ga0R (k ¢¢ , q , w )Taa
R
¢ (k ¢¢ , k ¢ , q , w )
(12.23)
and the quasi-statically screened potential is (with e being the

charge of the electron in vacuum)
4p e2 (12.24)
Waa ¢ (q ) = za za ¢
e b (q2 + k (q)2 )
where e b = nb2 , κ(q) is the inverse screening length, and za the sign
for the charge (–1 for electrons and +1 for holes). The unscreened
potential is V(q) = 4πe2/(εbq2). Details on how the effective potential
can be treated in a numerical solution of the equation set can be
found in Kwong et al. [14]. Note that the above set of equations
constitutes a self-consistency problem, as the self-energies depend
on the spectral function, but the spectral functions in turn depend
on the self-energies.
Once the luminescence spectral density is known, we can use the
cooling model developed by Sheik-Bahae and Epstein [7]. We define
the radiative recombination coefficient B (for ne = nh ∫ n ) via
dw
L(n) = B(n)n2 = Ú 2p R(w ) (12.25)
and use the rate equation for the excitation density evaluated under
stationary conditions,
dn a(wa , n)
= I - An - heL(n) - Cn3 = 0 (12.26)
dt wa pump
where Ipump is the pump intensity (more precisely irradiance), ωa
the frequency of the pump field, A the nonradiative recombination
coefficient, he the extraction efficiency of the luminescence (i.e.,
the fraction of the luminescence that leaves the crystal), and C the
Auger coefficient. Equation 12.26, together with the power balance
equation for the net power extracted from the system,
Pnet = Ipump ÈÎa(wa , n) + a b ˘˚ - w lheL(n) (12.27)

where αb is the parasitic background absorption coefficient and
w =
Ú dw w R(w ) (12.28)
Ú dw R(w )
the mean luminescence frequency, constitutes the basic framework
of the Sheik-Bahae–Epstein model.
The break-even nonradiative recombination coefficient Ab, which
marks the boundary where cooling becomes possible, is defined
by Pnet(Ab) = 0. Fixing all but two parameters, leaving only the
experimentally easily adjustable parameters ωa and ne as variables,
we optimize the break-even condition in the two-dimensional space
of ωa and ne, resulting in the optimized break-even nonradiative
coefficient
{
A ∫ max A(w a , ne ) P
net =0
} (12.29)
Figure 12.2 shows the results for the optimized breakeven non-
radiative lifetime (defined as the inverse of *A) as a function of
temperature. The solid and dotted lines show, respectively, the results
of the full calculation and the second-order approximation using the
free single-particle Green’s function G0 in the T matrix self-energies.
The overall temperature dependence is similar for these two curves
(the luminescence spectra in the full T matrix and second-order
approximations are quite different, but the resulting differences
in the values of the optimal density and radiative recombination
coefficient largely compensate each other, leading to similar values
for *τb), suggesting that many-body correlation effects beyond the
T matrix approximation would probably not change the results
significantly. In the so-called plasma theory model, the T matrix
self-energy is completely neglected, but the excitonic effects in the
susceptibility are still taken into account at a nonperturbative level.
In other words, in that model the exciton resonance is accounted
for in the susceptibility, but the many-particle processes that
renormalize the exciton resonance are reduced to pure electron–
hole plasma effects. We see that, qualitatively, plasma theory yields
results similar to the full theory. Only the model where all Coulomb
interactions are neglected, called “free” model, yields qualitatively
different results. We see from Fig. 12.2 that the existence of the
exciton resonance is the main factor that can make cooling at low
temperatures possible. More details can be found in Refs. [14–16].
Full
1000
2nd order
free
100 Plasma
10
(µs)
1
τb
0.1 Excitonic
Models
0.01
10 100 500
Temperature (K)
Figure 12.2 Optimized break-even nonradiative lifetime versus tempera-
ture for several different theories. For each model, the predict-
ed cooling region is above its curve. Reprinted from Ref. [16],
Copyright 2007, by the American Physical Society.
12.3 Effect of Luminescence Propagation and

Re-Absorption
In the previous section, we assumed the system to be a perfect
infinite crystal. Furthermore, we did not account for luminescence
re-absorption and imperfect luminescence extraction in the
microscopic theory. Rather, we included the luminescence extraction
efficiency he as a phenomenological parameter in the cooling model.
In the following, we generalize the theory to include light energy
transport, luminescence re-absorption, and surface or interface
reflection. A typical configuration including the semiconductor
sample and a transparent dome, intended to maximize the extraction
efficiency (see, for example, Sheik-Bahae and Epstein [7]) is shown
in Fig. 12.3.
Light energy transport theory is on the same footing as the
microscopic absorption theory. In addition to the electron Green’s
function, we now have a photon Green’s function D, which is
renormalized by the presence of the interband polarization.
Effect of Luminescence Propagation and Re-Absorption 387
Evaluated at equal times, the photon Green’s function yields the

photon Wigner function,
nqs (r , t ) = Â k
·cq - k /2,s cq + k /2,s Òei k ◊r (12.30)
where σ denotes the polarization state of the photon density, and

c(c*) c(c) are the photon annihilation (creation) operators. The
photon Wigner function determines the Poynting vector,
c c
Â
1
S(r , t ) = (E ¥ H ) = q w q Re(nqs (r , t ))
4p V qs
nb
(12.31)
= ÂÚ
qs
dw J q s w (r , t )
z
air

GaAs J q σ ω ( r ) d
ZnSe
Figure 12.3 Schematic of the geometry analyzed with the spectral energy
flux density. In the calculations, we assumed the lateral
extension to be infinite and as a consequence the net energy
flow to be restricted to positive and negative z directions
(thick blue vertical arrows). The extension in the z direction
was chosen to be 1 μm, and the partial or total reflection at the
GaAs/air and GaAs/ZnSe interfaces were taken into account
according to Fresnel’s equations.
In the last step, we defined the spectral density of the energy flux
Jqσω(r, t), which obeys the following transport equation:
Ï∂

Ó ∂ t
R ¸
( ) <
Ì + v q w q ◊ — - Im Pq w ˝ J q s w (r , t ) = - q Im Dq w i Pq w
˛
R
( )
(12.32)
Here, the second term in the parenthesis on the left-hand side
accounts for spatial propagation in the q direction with generalized
group velocity vqω, the third term accounts for luminescence re-
absorption, and the right-hand side represents the luminescence

R
source. The photon spectral function Dqw is usually taken to be a delta
function peaked at the free-photon dispersion, DqwR
ª d (w - w q ), with
ωq = cq/nb. We will come back to a more general photon spectral
function, which includes effects of excitons–polaritons, below.
The numerical solution of the photon transport equation allows
us to determine the escaped luminescence rate entering the rate
equation of the excitation density, Eq. 12.26,
Â ÂÚ
1 1
he B(n) n2 = dw z s ◊ J q s w ( zsurface ) t q s w ( zsurface )
dz q s
w q
(12.33)
surface
and the escaped luminescence energy entering the corresponding

rate equation for the net power, Eq. 12.27,
Â ÂÚ
1
dw z s ◊ J q s w ( zsurface ) t q s w ( zsurface )
w l he B(n) n2 =
dz qs
(12.34)
surface
Here, tqσω(Zsurface) is the transmission coefficient according

to Fresnel’s equations. With the solution of the photon transport
equation, we can now predict the value of the extraction efficiency.
In addition to the important effects of luminescence re-
absorption and interface reflection, we have also addressed the
issue of exciton–polariton propagation. It is frequently assumed
that exciton–polariton effects play a negligible role in the cooling
theory. In the following we investigate this issue within the limits of
photon Wigner function theory. Within this theory, polariton effects
enter locally; at each point in space the retarded photon Green’s
function DqRw (r ) can exhibit anti-crossing between exciton states
and photon states. An example is shown in Fig. 12.4 for negligible
excitation density and a very low temperature of T = 5 K. We chose
a temperature this low in order to demonstrate the principle
existence of polariton effects (anticrossing). At higher temperatures
(we usually investigate cooling between 10 K and 500 K) and higher
densities, this effects is practically absent, because the exciton
linewidth exceeds the longitudinal-transverse splitting (the relevant
energy scale for the anticrossing).
e-h polariton
w
2s
1s
q
Figure 12.4 Exciton–polariton dispersion at a temperature of 5 K, as seen
in the peaks of the logarithm of the photon spectral function
[-2ImDqw R
]. The anticrossing between the 1s exciton and the
photon dispersion is clearly visible, and to a lesser extent even
that of the 2s exciton and the photons. In the e–h continuum,
the polariton dispersion is dominated by that of free photons.
The (quadratic) spatial dispersion of the 1s and 2s excitons
cannot be seen in this figure because the chosen q range is very
small.
In our photon transport equation model, the boundary conditions

determining the interface reflection and transmission are solved
separately for each (ω, q) value (which is an approximation to the
proper boundary conditions). One can assume that luminescence
from the lower polariton branch is suppressed at large wavevectors.
As far as escaped luminescence is concerned, exactly this happens in
our model, as can be seen by comparing Figs. 12.5 and 12.6. We see
clearly that the luminescence observed outside the semiconductor
is insensitive to polariton effects, even at temperatures as low as 5 K
and at negligible excitation density. An effect much more important
is the luminescence re-absorption, which is also shown for the
example of T = 5 K in Fig. 12.6.
1010
Luminescence
106
102
-1.0 -0.5 0.0
(ω- ER) / Eg
Figure 12.5 Photoluminescence spectra at T = 5 K as hypothetically seen
inside the semiconductor with (red curve) and without (blue
curve) polariton effects. The polariton luminescence contains
a large component from the lower polariton branch (at large
wavevectors), which however are not transmitted to the
outside (see Fig. 12.6).
1010
Luminescence
106
102
-1.0 -0.5 0.0
(ω - ER)/Eg
Figure 12.6 Photoluminescence spectra at T = 5 K as seen outside the
semiconductor with (red solid curve) and without (blue dashed
curve) polariton effects (these two curves are practically
indistinguishable). The green curve shows results neglecting
luminescence re-absorption.
To further extend our analysis, we combine now our theory of

luminescence from p-doped semiconductors, see Rupper et al. [22,
23], with the light transport theory outlined above. We formulate the
theory of luminescence from a partially ionized exciton gas with the
additional luminescence channel between acceptors and conduction
band electrons. Including this channel, the spectrally integrated
escaped luminescence can be written as
Loutside = ηeB(ne)ne(ne + Na) (12.35)
and the corresponding spectrally integrated escaped luminescence
energy is
ħωlηeB(ne)ne(ne + Na) (12.36)
Here, Na is the acceptor density and ne the optically excited
electron density. Using the numerical solution for the spectral
energy flux density, we can compute the extraction efficiency and
the minority carrier lifetime, the latter being defined as τ–1 ≈ Loutside/
ne.
Figure 12.7 shows the calculated extraction efficiency. The
decrease of ηe with decreasing temperature is expected and can be
attributed to the spectral sharpening of the exciton resonance, which
in turn leads to an increase of luminescence re-absorption. The
sharp rise of ηe at temperatures below 30 K may seem surprising. It
can be explained by the onset of electron-to-acceptor luminescence
(acceptor freeze-out), because this luminescence is spectrally below
the exciton resonance and thus suffers less from re-absorption.
Loutside
ηe =
0.15 Llocal
ηe
0.10
N a = 3 ×1015 cm −3
0.05
0 100 200 300 400 500

Temperature (K)
Figure 12.7 Calculated luminescence extraction efficiency for a p-doped
structure.
Figure 12.8 shows calculated results for the minority carrier

lifetime. These results are in good agreement with the experimental
findings from the University of New Mexico team (see Imangholi et
al. [11]). Their measurements of the lifetime τ yield the same general
trends as a function of temperature (see Fig. 12.8). For example at
T = 500 K, the measured value is 8.3 μs, which is very close to our
value calculated for a similar structure.
10-4
N a = 3 ×1015 cm −3
ZnSe dome structure
10-5
Lifetime (s)
t t = 8µs
-6
10
10-7 he t
10-8
0 100 200 300 400 500
Temperature (K)
Figure 12.8 Calculated minority carrier lifetime for a p-doped GaAs
structure with a ZnSe dome.
12.4 Finite Spatial Beam Profiles

In the following we study the effect of radial spatial photon transport
(see also Rupper et al. [24]). In typical experiments, the beam profile
is smaller than the extension of the semiconductor sample. For
example, the beam profile can be Gaussian (with azimuthal symmetry)
and the sample can be disk-like. Typical numbers for the beam waist
are 100 μm with a disk diameter of 1000 μm or larger (see Fig.12.9).
The cooling process is sensitive to the carrier density. If the density
is too large, then Auger recombination and/or band filling prevents
cooling. If the density is too low, then nonradiative recombination
Finite Spatial Beam Profiles 393
prevents cooling. As such, the spatially inhomogeneous optical

excitation can be expected to have negative consequences for the
cooling process. Since the excitation power drops away from the
beam center, the cooling threshold can be expected to be reached
only in a small fraction within the beam. Hence it would be possible
that the sample cools in that narrow region, while it heats in the
beam’s radial wing. This effect is expected to be exacerbated by the
re-absorption of the luminescence propagating (laterally) inside
the sample. As is schematically shown Fig. 12.10, a fraction ηe (the
extraction efficiency) of the total luminescence generated at a given
position inside the sample is able to escape from the sample. The
remainder, that is, the fraction 1 – ηe of the luminescence, propagates
laterally inside the sample. If we assume the sample’s lateral
extension to be much larger than the absorption length (=inverse
absorption coefficient) of the emitted light, the laterally propagating
luminescence light will be completely re-absorbed. Depending on
the specific spatial profile of the re-absorption, this process can also
be expected to hinder cooling. Also, lateral charge carrier diffusion
can affect the spatial density and power profile, and thus influence
the cooling process, but we have not yet addressed this issue in our
studies.
light
beam
r2
−
I pump (r ) = I 0 e 2σ 2
spatial profile
of excitation
Figure 12.9 Sketch of excitation geometry with a disk-like GaAs sample

and a Gaussian beam. Reprinted with permission from Ref.
Engineers.
Lout (r ) = ηe L(r )
Lin (r ) = (1 − ηe ) L(r )
out in
r
0
Figure 12.10 (Top sketch) Sketch of luminescence propagation. A fraction
ηe of the luminescence produced at a given position r in the
sample escapes from the sample, whereas the rest (1 – ηe)
propagates laterally in the sample and is re-absorbed for
sufficiently laterally extended samples. (Bottom sketch) In our
model, the re-absorption at position r has two contributions
from lateral photon propagation, one from luminescence
propagation laterally outward and the other laterally inward.
Reprinted with permission from Ref. [24], Copyright 2009,
Society of Photo Optical Instrumentation Engineers.
We assume that at each point in the structure, the fraction 1

– ηe of the locally generated luminescence propagates inside the
medium. For simplicity, we assume the propagation direction to be
restricted to the radial direction, either outwards (away from the
center) or inwards (toward the center) (see Fig. 10, bottom). We
are using the model for photon transport discussed above, which is
based on the Poynting vector (Eq. 12.31). To simplify calculations,
we adopt a model in which photons are transported only along
the radial direction (in cylindrical coordinates), that is, Jq, σ,ω(r) is

nonzero only when q e r (outward flow) or q -e r (inward flow),
where e r is the radial unit vector. We write
J q,s ,w (r ) = d (qz ) ÎÈ Jqout ˘
,s ,w (r )d (fq - f ) - J q,s ,w (r )d (fq - f - p )˚ e r
in
(12.37)
(where fq is the azimuthal angle of q). The transport equation,
including the luminescence source term and the luminescence re-
absorption term, with α(r, ω) ∫ α (n(r), ω), reads
Finite Spatial Beam Profiles 395
Ï 1 ∂ ¸
Ì± r + a (r ,w )˝ Jqout/in
,s ,w (r ) = sq,w ,s (r ) (12.38)
Ó r ∂ r ˛
where the upper sign is for Jout and the lower sign for Jin, respectively.
Using the boundary conditions Jq,s ,w ( • ) = 0 and
in
Jq,s ,w (0) = Jq,s ,w (0) (which accounts for the fact that inward flow
out in
passes through the origin and then contributes to outward flow), we

find the outgoing flux to be
• Ê r
- Ú dr ¢¢a ( r ¢¢ ,w )
r
- Ú dr ¢¢a (|r ¢¢|,w )
ˆ
Á ˜
Ú
1
Jqs w (r ) =
out
dr ¢r ¢ s(r ¢ , q ,w )Á q(r - r ¢ )e r ¢ + e -r ¢ ˜
r
0 ÁË ˜¯
(12.39)
and the incoming flux is
r¢
•
Ú
- dr ¢¢a ( r ¢¢ , w )
Ú
in 1
Jqs w (r ) = dr ¢r ¢ s(r ¢ , q , w ) e (12.40)
r
r
r
The luminescence source sq,ω,σ(r) is normalized such that the total
in-plane (lateral) luminescence (i.e., sq,ω,σ(r) summed/integrated
over all wavevectors, frequencies, and polarization states) yields the
fraction 1 – he of the globally generated luminescence (integrated
over all space). We have modified the rate equations entering
the Sheik-Bahae–Epstein model by including the re-absorption
calculated with the lateral photon transport model. The modified
rate equation for the density profile reads
a (w a , n(r ))
Ipump (r ) - An(r ) - B (n(r )) n(r )2 - Cn(r )3 + Qn (r )
w a

dn(r )
= =0
dt
(12.41)
Note that the luminescence term (Bn2) is not multiplied by the
extraction efficiency. Instead, we include a source term that accounts
for the re-absorption of carriers due to lateral luminescence
propagation, which is given in terms of the photon flux and the
absorption coefficient,
Qn (r ) = Â a (r ,w ) | J
q,s ,w
q,s ,w (r ) | (12.42)
Similarly, we have for the power balance equation
Pnet (r ) = Ipump (r ) ÈÎÍa (w a , n(r ) ) + a b ˘˚˙

(12.43)
- w (r )B (n(r )) n(r )2 + Qe (r )
with the source term due to luminescence re-absorption
Qe (r ) = Â a(r,w )w | J
q,s ,w
q,s ,w (r ) | (12.44)
Figures 12.11 and 12.12 show results with and without the lateral
photon transport included. We see that, apart from slight numerical
noise, the results for the power is very similar in both cases,
implying that lateral photon transport under typical experimental
conditions should not be expected to prohibit cooling. As expected,
the density profile in the case with lateral photon transport is
smeared out beyond the region of optical excitation by the external
light beam. However, the off-axis density produced by luminescence
re-absorption is much smaller than the density excited directly by
the external beam. This explains why the power profiles are very
similar in the two cases (with and without lateral photon transport).
Without energy flow τnr=0.4µs

1018
Without energy flow τnr=16µs
With energy flow τnr=0.4µs
1017 With energy flow τnr=16µs
Density (cm-3)
1016
1015
1014
0 100 200 300 400 500
Distance from center (µm)
Figure 12.11 Spatial density profile for excitation with a Gaussian beam
(σ = 100 μm) at temperature T = 300 K with spatial photon
transport. Reprinted with permission from Ref. [24], Copyright
2009, Society of Photo Optical Instrumentation Engineers.
Theory of Passivation Layers 397
2x105
Net Power (W/cm3)

0
-2x105 Without energy flow τnr=0.4µs

Without energy flow τnr=16µs
With energy flow τnr=0.4µs
-4x105 With energy flow τnr=16µs
-6x105
0 100 200 300 400 500
Figure 12.12 Spatial profile of the net power for excitation with a Gaussian
beam (σ = 100 μm) at temperature T = 300 K with spatial
photon transport. Reprinted with permission from Ref. [24],
Engineers.
12.5 Theory of Passivation Layers

So far, our theoretical descriptions treated only spatially homoge-
neous semiconductors, albeit under possibly inhomogeneous exci-
tation conditions. In typical experiments, however, semiconductors
are often heterostructures designed to reduce nonradiative recombi-
nation at the sample’s surface. In particular, GaAs/GaInP structures
have been used. In these structures, the GaAs layers are usually un-
intentionally p-doped, while the surface layers of GaInP are n-doped.
The spatial inhomogeneties that we address in the following
section are those related to the doping profile of the semiconductor
in the growth direction. We call these longitudinal inhomogeneties
in order to distinguish them from the lateral spatial inhomogeneties
discussed in the previous section. The structures used in the
University of New Mexico experiments are doped structures, and
many of them are of the form GaInP/GaAs/GaInP (see Fig. 12.13).
The role of the GaInP layers is to minimize nonradiative surface
recombination and hence act as a passivation layer. A critical question
related to this structure concerns the spatial separation of electrons
and holes. It had been known for some time that the band profile of
this structure is of type II character, which means that the conduction
band has a maximum in the GaAs region. In that case, the electrons
tend to accumulate in the energetically more favorable GaInP layers,
away from the holes that remain in the GaAs layer. Such a spatial
charge separation would hinder photo-luminescence and hence
could be detrimental for optical refrigeration. To address this issue,
we first determine the spatial band profile and electron and hole
density profiles across the n-p-n structures. With that knowledge,
we use our microscopic photoluminescence and absorption theory
to determine the spatial variation in the luminescence efficiency and
the absorption spectra.
GaInP GaAs GaInP

n-doped p-doped n-doped
Si C or Zn Si
1017 cm-3 3x1016 cm-3 1017 cm-3
n p n
Figure 12.13 Schematic of the doping profile of the structure, which was used
in the University of New Mexico experiments, see Imangholi et
al. [11], with typical doping concentrations.
The theory for the band and density is an extension of the

conventional textbook treatment of heterogeneous and doped
semiconductor structures via self-consistent solutions for the
Poisson equation and the carrier density profiles. While the
conventional textbook treatment is restricted to structures in
thermal equilibrium, in which charge carries can only originate
from the doping, our generalization allows us to include optically
excited carriers (Rupper et al. [22, 23, 25]). The fundamental set
of equations consists of the space-dependent acceptor and donor
occupation functions
Da
fa ( z ) =
Da + e
(
b e a + eV( z )- mh - e 0v ( z )) (12.45)
and
Dd
fd ( z ) = (12.46)
b ( e d -eV ( z )- me - e c0 ( z ))
Dd + e
where V(z) is the self-consistently computed electric potential, β

= 1/kBT the inverse temperature, εa the acceptor and εd the donor
energy, Da/d the degeneracies of the acceptor and donor levels,
respectively, and e v/c
0
( z ) the energy band profiles of the separated
layers. The free electron density in the conduction band is given by
the Fermi functions,
ne ( z ) = 2Â
1
(12.47)
1+e

k
(
b e e (k )- eV ( z )- me - e c0 ( z ) )
and similarly for the valence band holes,
nh ( z ) = 2 Â
1
(12.48)
b (e (k )+ eV ( z )- mh - e v0 ( z ))
1+e h
k
The electric field in the GaAs layer is

4p e Ê z zj ˆ
E( z ) =
eGaAs ÁË Úzj
dz ¢ ncharge ( z ¢ ) + Ú ze
dz ¢ ncharge ( z ¢ )˜ (12.49)
¯
and in the GaInP layer it is
4p e z
E( z ) =
eGaInP Ú Ze
dz ¢ ncharge ( z ¢ ) (12.50)
This is determined by the charge density

ncharge ( z ) = nh ( z ) + Nd+ ( z ) - ne ( z ) - Na- ( z ) (12.51)
with
Ú dz n charge ( z ) = 0 (12.52)
where the concentration of ionized donors Nd+ ( z ) follows from the

donor occupation function and similarly Na- ( z ) from the acceptor
occupation function. Finally, the electric potential is given by
z
V (z) = - Ú 0
dz ¢ E( z ¢ ) (12.53)
Figure 12.14 shows a flowchart of the numerical solution

algorithm.
Figure 12.15 shows the band profile without the electric potential.
It shows that the structure could be called a weak type I structure,
in the sense that the potential wells for electron and holes are lined
up (they are both in the middle layer), and thus no charge separation
would be expected. However, the potential for the electrons is rather

shallow, in that sense the type I characteristic could be called weak.
Including the effects of the self-consistent electric potential, we
see in Fig. 12.16 that a structure in thermal equilibrium is clearly
a type II structure, in the sense the electrons and holes reside at
different spatial locations. In this case, the electrons experience a
potential barrier in the middle layer, and are thus pushed to the side
layers, while the holes see a well in the middle layer, thus residing
there. The situation changes when we consider an optically excited
structure, such as occurs under optical refrigeration conditions. As
we see from Fig. 12.16, optical excitation tends to flatten the bands,
in this case especially the conduction band, which means that spatial
charge separation due to a type II characteristic is not expected to
prevail under optical excitation conditions.
Initialize V(z)
calculate µe using ne(z=0)
electron density and ionized donor profile
calculate µh from overall charge neutrality
hole density and ionized acceptor profile
calculate electric field and potential V(z)

Figure 12.14 Schematic of the self-consistent solution for the n-p-n
structure with the possibility of optically excited carrier for
given temperature and center electron density ne(z = 0).
One of the main objective in using an structure with n-doped

layers on each side is the reduction of non-radiative recombination.
This type of recombination, which inhibits optical refrigeration,
is often dominated by surface and interface recombination. The
presence of the n-doped passivation layers almost removes the

effect of nonradiative recombination at the surface (in this case
the outside surface of the n-doped GaInP layers. There remains
recombination taking place at the GaAs–GaInP interface, and in
the following we discuss the properties of those recombination
processes and possible control thereof. The control of nonradiative
interface recombination is based on the general idea of keeping
the minority carrier concentration in the vicinity of the interface
as low as possible. Figure 12.17 shows a sketch of the electron and
hole concentration and clarifies that, trivially, holes (electrons) are
minority carriers in the n-doped (p-doped) segments. What is not
immediately clear from this sketch is whether electrons or holes are
minority carriers at the interface.
1.5 Conduction band

Energy (eV)
1.0
1.4224 eV
1.91 eV
0.5
0.0
-0.5
Valence band
-1.0 -0.5 0.0 0.5 1.0

Figure 12.15 Energy band profile. The lower curve is for the valence band
(hole density) and the upper curve for the conduction band
(electron density). Reprinted with permission from Ref. [23],
Copyright 2010, AIP Publishing LLC.
A simple model for nonradiative interface recombination is

based on Shockley–Read–Hall processes:
∂ne ( z ) ∂n ( z ) ne ( z j )nh ( z j )
∂t nr
= h
∂t nr
=- Â
j =1 ,2
d(z - z j ) S
ne ( z j ) + nh ( z j )

~- d ( z - z j ) S nmin ( z j )
(12.54) Â
j =1 ,2
1.5
1.0
Energy (eV)
0.5
0.0
-0.5
-1.0 -0.5 0.0 0.5 1.0

Figure 12.16 Band profile for conduction (upper green curves) and valence
(lower red curves) bands. Thermal equilibrium at T = 300
K is shown as thin lines. The type II character (electrons in
conduction band pushed away from GaAs region and holes in
valence band localized in GaAs region) is apparent. The thick
curves show an optically excited structure with an electron
density at the center of ne (z = 0) = 1017 cm–3. The optical
excitation has led to flat-band conditions for the conduction
band.
- - - - - - - - - - - - - - - - - - -
n p n
+ + ++++ ++++ + +
zc zj ze
Figure 12.17 Sketch of n-p-n structure with electrons and holes schematically
indicated.
where S denotes the surface recombination velocity and

nmin(zj) = min{ne(zj), nh(zj)} (12.55)
is the minority carrier density at interface. Assuming that the
optically excited carrier density naopt ( z ) is much larger than the
intrinsic carrier density, we have
∂neopt ( z ) ∂nhopt ( z )

∂t
=
∂t
=- Â
j =1 ,2
d ( z - z j ) S nmin ( z j ) (12.56)
nr nr
1016
Density (cm-3)
1014
1012 1015 cm-3
1010
-1.0 -0.5 0.0 0.5 1.0
Figure 12.18 Electron (solid green line) and hole (dashed red line) density
profile at T = 300 K and with an acceptor density of Na = 3.5
× 1015 cm–3. The optical excitation is chosen to yield a carrier
density at the center of 1015 cm–3. Holes are majority carriers
at the center but minority carriers at the interface.
We introduce the average carrier density (averaged over the

GaAs layer of thickness d)
1
na = Ú dz n a ( z ) (12.57)
opt
d
and assume the layer thickness to be sufficiently large so that the
average density is the same as the density at the center of the layer,
na ª naopt ( zc ) for a = e or h, we have
∂neopt ( z c ) ∂nhopt ( z c ) 2S
= =- n (z )
∂t ∂t d min j
nr nr
2S nmin ( z j )
=- n ( z ) (12.58)
d nmin ( zc ) min c
= - A nmin ( zc )
where we introduced an effective nonradiative recombination

coefficient A, whose inverse is the effective nonradiative decay time
1 d nmin ( zc )
t nr = = (12.59)
A 2S nmin ( z j )
This shows that, as mentioned above, improving (lengthening)

the nonradiative lifetime can be achieved by minimizing the minority
carrier density at the interface, nmin(zj).
100
nh(zj) / nmin(zc)
10-2
10-4
10-6 10
10 1012 1014 1016 1018
nmin(zc)
Figure 12.19 Dependence of the hole density at the interface versus minority
center density. The solid line is for Na = 1014 cm–3, Nd = 1017
cm–3, the dash-dotted line for Na = 1014 cm–3, Nd = 1019 cm–3,
and the dashed line for Na = 3.5 × 1017 cm–3, Nd = 1017 cm–3. The
thickness d is taken to be 1 μm and the recombination velocity
S = 500 cm/s.
Figure 12.19 shows the dependence of the interface hole density

as a function of the optical excitation, here parametrized by the
optically excited density at the center, or minority center density
nmin(zc). With increasing nmin(zc), the interface hole density increases.
Since the holes are the minority carriers at the interface, this in
turn means that the effective nonradiative decay time decreases,
as is shown in Fig. 12.20. However, an increased n-doping level, in
our example of Nd = 1019 cm–3, is found to decrease the interface
hole density (Fig. 12.19) and hence improve (increase) the effective
nonradiative lifetime (Fig. 12.20). As shown in Fig. 12.20, such an
increased n-doping concentration can, for the idealized model of
and infinitely thin interface region, yield lifetimes comparable to
the optimized break-even lifetimes discussed above. This, however,
assumes that a high n-doping concentration does not open up new
nonradiative transitions that would again decrease the effective
nonradiative lifetime.
106
n
104
(µs)
102
τnr
100 τb
10-2 10
10 1012 1014 1016 1018
nmin(zc)
Figure 12.20 Effective nonradiative recombination time as a function of
center density. The three different thick line styles correspond
to the same doping levels as in Fig. 12.19. The vertical and
horizontal dotted lines indicate the optimized cooling density
and optimized break-even lifetime, respectively, obtained from
the model of bulk semiconductors discussed above in Section
12.2.
Finally, we show in Fig. 12.21 the cooling efficiency as a function

of minority center density. We see that positive cooling efficiency can
be achieved for moderate minority center densities (in other words,
optical pumping that yields densities small enough so that Auger
recombination does not become dominant), and high n-doping
concentrations in the passivation layers. Further details can be
found in Rupper et al. [23].
0.0
ηc
-0.1
-0.2 10
10 1013 1016 1019
Minority center density (cm –3)
Figure 12.21 Cooling efficiency versus minority center density. The three
different line styles correspond to the same doping levels as
in Fig. 12.19. The dash-dotted green line shows that cooling
should be possible, hc > 0 , for low to moderate minority center
densities (i.e., moderate optical pumping powers) and large
n-doping concentration in the passivation layers. Reprinted
with permission from Ref. [23], Copyright 2010, AIP Publishing
LLC.
Acknowledgments
We thank M. Sheik-Bahae, B. Imangholi, M. Hasselbeck, R. Epstein,
and J. Khurgin for helpful discussions. We acknowledge financial
support from AFOSR MURI grant no. A9550-04-1-0356 and addi-
tional support from JSOP and TRIF Photonics.
References
1. Pringsheim, P. (1929). Zwei Bemerkungenueber über den Unterschied

von Lumineszenz- und Temperaturstrahlung, Z. Phys., p. 739.
2. Epstein, R. I., Buchwald, M. L., Edwards, B. C., Gosnell, T. R., Mungan,
C. E. (1995). Observation of laser-induced fluorescent cooling of a
solid, Nature, 377, pp. 500–503.
References 407

Phys., 73, p. 086501.
5. Oraevsky, A. (1996). Cooling of semiconductors by laser radiation, J.
Russ. Laser Res., pp. 471–479.
6. Rivlinand, L., Zadernovsky, A. (1997). Laser cooling of semiconductors,
Opt. Commun., 139, pp. 219–222.
7. Sheik-Bahae, M., Epstein, R. I. (2004). Can laser light cool semiconduc-
tors?, Phys. Rev. Lett., 92, p. 247403.
8. Apostolova, T., Huang, D., Alsing, P. M., Cardimona, D. A. (2005).
Comparison of laser cooling of the lattice of wide-band-gap
semiconductors using nonlinear or linear optical excitations, Phys. Rev.
A, 71, p. 013810.
9. Khurgin, J. (2007). Surface plasmon-assisted laser cooling of solids,
Phys. Rev. Lett., 98, p. 177401.
10. Gauck, H., Gfroerer, T. H., Renn, M. J., Cornell, E. A., Bertness, K. A.
(1997). External radiative quantum efficiency of 96 percent from a
GaAs/GaInP heterostructure, Appl. Phys. A, 64, pp. 143–147.
11. Imangholi, B., Hasselbeck, M. P., Sheik-Bahae, M., Epstein, R. I., Kurtz,
S. (2005). Effects of epitaxial liftoff on interface recombination and
laser cooling in GaInP/GaAs heterostructures, Appl. Phys. Lett., 86, p.
081104.
14. Kwong, N. H., Rupper, G., Binder, R. (2009). Self-consistent T-matrix
theory of semiconductor light-absorption and luminescence, Phys. Rev.
B, 79, p. 155205.
15. Rupper, G., Kwong, N. H., Binder, R. (2006). Large excitonic enhancement
of optical refrigeration in semiconductors, Phys. Rev. Lett., 97, p.
117401.
16. Rupper, G., Kwong, N. H., Binder, R. (2007). Optical refrigeration of
GaAs: theoretical study, Phys. Rev. B, 76, p. 245203.
17. Zimmermann, R. (1987). Many-Particle Theory of Highly Excited
Semiconductors (Teubner, Germany).
18. Kwong, N. H., Rupper, G., Gu, B., Binder, R. (2007). The relation between
light absorption and luminescence in laser cooling of two-dimensional
semiconductor systems, Proc. SPIE, 6461, p. 64610I.
19. Gu, B., Kwong, N. H., Binder, R. (2013). Relation between the interband
dipole and momentum matrix elements in semiconductors, Phys. Rev.
B, 87, p. 125301.
20. Schmielau, T., Manske, G., Tamme, D., Henneberger, K. (2000).
T-matrix approach to the linear optical response of highly excited
semiconductors, Phys. Status Solidi B, 221, pp. 215–219.
21. Piermarocchi, C., Tassone, F. (2001). Role of bound pairs in the optical
properties of highly excited semiconductors: a self-consistent ladder
approximation approach, Phys. Rev. B, 63, p. 245308.
22. Rupper, G., Kwong, N. H., Binder, R. (2008). Theory of luminescence
and optical refrigeration in p-doped semiconductors, Proc. SPIE, 6907,
p. 690705.
23. Rupper, G., Kwong, N. H., Binder, R., Li, Ch.-Y., Sheik-Bahae, M. (2010).
Effect of n-p-n heterostructures on interface recombination and
semiconductor laser cooling, J. Appl. Phys., 108, p. 113118.
24. Rupper, G., Kwong, N. H., Binder, R. (2009). The role of finite spatial
beam profiles on photo-luminescence and laser cooling in GaAs
structures, Proc. SPIE, 7228, p. 722805-1.
25. Rupper, G., Kwong, N. H., Binder, R. (2010). Theory of time-resolved
photo-luminescence and carrier lifetime measurements in GaAs/
GaInP heterostructures, Proc. SPIE, 7614, p. 76140D-1.
Chapter 13
Coulomb-Assisted Laser Cooling of

Piezoelectric Semiconductors
Iman Hassani Nia and Hooman Mohseni

EECS Department, Northwestern University, 2145 Sheridan Rd., M250,
Tech Building Evanston, IL 60208, USA
hmohseni@northwestern.edu
13.1 Introduction
Cooling materials with light is a topic that has attracted tremendous
interest in recent years. Researchers usually find the theory
counterintuitive, and, in fact, it was a highly disputed idea when
Pringsheim first proposed it in 1929 [35]. However, only 66 years
later, the first observation of laser cooling of solids was made by
Epstein in 1995 in trivalent ytterbium-doped heavy metal fluoride
glass [12]. In rare earth ions, such as ytterbium, the electronic 4f
levels are shielded from their surroundings by the filled 5s and 5p
shells, leading to suppression of multiphonon relaxation and a high
radiative recombination efficiency favorable for optical refrigeration
[36]. Since the first discovery of laser cooling, a methodical

www.panstanford.com
410 Coulomb-Assisted Laser Cooling of Piezoelectric Semiconductors
approach by many scientist led to a significant improvement of

optical refrigeration starting from 0.3° in 1995 by Epstein to more
than 193° in 2014 [25]. Such significant progress was the result of
optimized purification and doping concentrations of the rare earth
ions in the host material. Laser cooling in rare earth–doped systems
are considered in Chapters 2–4 of this book.
It should be noted that the practical laser cooling method of
solids is mainly based on the anti-Stokes mechanism [29–31], which
is schematically depicted in Fig. 13.1. The absorption of a laser in
the material creates electron–hole pairs, and the photogenerated
carriers go through different scattering and recombination
processes before reaching thermal equilibrium. Now suppose that
the carriers are excited with energies close to the band edge and,
after reaching thermal equilibrium, they absorb a net energy from
the lattice vibrations (which are called phonons). As a result, the
thermalized carriers have more energy than the absorbed photons’
energy. The energy by which the carriers exceed the photon’s energy
is roughly equal to kT (where k is the Boltzmann constant and T is
the temperature). Consequently, the photoluminescence (PL) of the
thermalized carriers has a higher energy than the absorbed laser.
It is therefore expected that a full extraction of the PL would be
equivalent to extraction of the thermal energy from the material to
the outside environment by means of radiation and the material cools
down. This is quite important because operating the optoelectronic
devices at lower temperatures usually reduces their electrical noise
level [13, 14].
kT
Input Laser
Photo -
Luminescence
kT
Figure 13.1 The process of anti-Stokes laser cooling of semiconductors.
Introduction 411
In practice, there are challenges that can forbid laser cooling

of solids by the above simple picture. One important issue is
the existence of nonradiative recombination pathways that
turn all of the absorbed energy into heat. For semiconductors,
the nonradiative recombinations can be divided into two main
categories: (1) the Shockley–Reed–Hall (SRH) recombination and
(2) the Auger recombination process. The SRH process is mainly
due to the interaction of the carriers with crystal nonidealities
such as surface states and defects. On the other hand, the Auger
recombination arises from the Coulomb interaction and scattering
of three carriers; the interaction between two carriers gives energy
to the third carrier and scatters it to a state with a higher energy.
The third carrier eventually relaxes to a lower energy, losing its
excess energy to the phonon bath. To realize laser cooling, both of
the aforementioned processes should be suppressed. While the
SRH process can be suppressed by increasing the material quality,
surface passivation and preventing the carriers from reaching the
defective area, the Auger recombination can be suppressed by
appropriate band structure engineering [27] and operating at lower
laser powers to avoid overly high photogenerated carrier densities.
It should be noted that the ratio of the radiative to the nonradiative
recombination rates increases as the operating temperature
decreases. But unfortunately, at the same time, the thermal energy
gained by the carriers (approximately equal to kT) decreases.
Another important challenge is to extract the PL out of the
material. For the case of rare earth–doped glass, the refractive index
of the host crystal is close to air (for glass it is approximately 1.5).
However, the difference between the index of the semiconductors
(around 3) and air is high. As a result, the issue of PL trapping arises
in semiconductors. The PL trapping results in photon recycling
and eventually wastes some portion of the luminescent energy to
nonradiative and heat-producing mechanisms. For this reason,
effective luminescence extraction methods need to be developed for
materials with a higher index of refraction.
To have a better understanding of the above explanations, we
note that the cooling efficiency (ηc) can be calculated as [41]:
E pl
hc = habshex -1 (13.1)
Ei
where Epl is the average energy of the PL, Ei is the average energy
of the incident photons, and ηex and ηabs are the external quantum
efficiency and the absorption efficiency, respectively. The external
quantum efficiency (ηex) can be calculated as [42]:
he Bn2
hex = (13.2)
An + he Bn2 + Cn3
where ηe is the extraction efficiency [28], n is the photogenerated
carrier density, and A, B, and C are the SRH, radiative, and Auger
recombination coefficients, respectively. This parameter is a measure
of the fraction of luminescence power extracted out of the material
before getting reabsorbed. The absorption efficiency is equal to:
a
habs = (13.3)
a + ab
where α is the absorption in the active region and αb is the background
absorption due to free carriers, impurities, and the defects of the
material. It is straightforward to conclude from Eqs. 13.2 and
13.3 that in order to maximize the net cooling efficiency, both the
external quantum efficiency and the absorption efficiency should
be maximized. The external quantum efficiency is dependent on the
material quality, the extraction efficiency, and the carrier density.
The extraction efficiency can be increased to almost 100% using
novel techniques that suppress both the total internal reflection
and the Fresnel reflection inside a high-index material. However,
the material quality sets a limit on maximum attainable external
quantum efficiency and absorption efficiency. The materials of choice
for semiconductor laser cooling up to now are GaAs [42] and CdS
[51]. Considering the bandgaps of these materials, the maximum net
cooling efficiency is approximately 2% for gallium arsenide and 1%
for cadmium sulfide at room temperature. These values are quite
low, rendering the laser cooling of bulk GaAs impossible, and for CdS,
only a particular nanoribbon structure has been cooled using a laser.
This is due to the insufficient external and absorption efficiencies
of these materials. These challenges suggest that mechanisms other
than anti-Stokes should be developed and employed in order to have
practical laser cryocoolers in the future. While different methods
of laser cooling for gaseous systems (such as Doppler cooling,
Sisyphus, resolved-sideband cooling, and sympathetic cooling) have
Piezoelectricity in Semiconductors 413
been implemented, the laser cooling of solids has remained a vast

area to explore new ideas. In the following sections, we start by
describing different types of Coulomb interactions in semiconductor
quantum wells (QWs), which have the potential for optical
refrigeration. The approach that we are going to explain is based
on the Coulomb interaction between the photogenerated electron
and holes in piezoelectric QW structures. For this reason, the next
section is devoted to the topic of piezoelectricity in semiconductors
and its previous application. We also discuss the effects of Coulomb
interactions in semiconductors in Section 13.4. This provides the
essential basis for understanding laser cooling of piezoelectric
semiconductors, which is explained in Section 13.2. We finish the
chapter with our concluding remarks about limitations and the
possible improvement of the described systems in Section 13.8.
13.2 Piezoelectricity in Semiconductors

The piezoelectric effect refers to the creation of polarized charge
by applying stress in a material that contains more than one type
of atom without geometrical inversion symmetry. To begin with,
we consider zinc-blende semiconductors, which are the simplest
semiconductor crystals for the investigation of piezoelectricity.
The induced polarization (P) can be related to stress (S) using the
piezoelectric tensor (e) [2]:
P = eS (13.4)
In most solids, the stress can be characterized using only six
components (Sxx, Syy, Szz, Sxy, Syz, Sxz) , which constitute the elements of
the stress vector S. The first three components (Siii = x,y,z) are called
biaxial tensor components, whereas the other three components
(Siji,j = x,y,z, i ≠ j) are the shear stress components. The applied
stress results in deformation of the crystal, and the atoms deviate
from their equilibrium positions. The stress (S) and the strain (ε)
of a crystal are connected by Hooke’s law and through the elasticity
tensor (C) [5, 44]:
S = Cε (13.5)
Similar to the stress, the strain of the material is also comprised
of six components. In Voigt notation they are defined as
∂ua
a , b = , u = u,x u y , uz x b = x , y , z (13.6)
∂x b a
where uα is the lattice displacement along the direction specified by
xα. The strain tensor is symmetric and the following relation holds
[50]:
∂ua ∂ub
ab =  ba = + (13.7)
∂x b ∂xa
Equations 13.4–13.7 provide a complete description of
piezoelectric field as a result of different types of stress and strain in
the material. It is important to note that because of crystal symmetry,
many elements of the elastic constants and the piezoelectric tensor
are essentially zero. For example, for a cubic crystal, the values of the
uniaxial elastic constants along the three axis are the same because
of spatial invariance (Cii=Cjj, i ≠ j). A similar relation holds for shear
components. Furthermore, shear strain cannot produce a normal
stress (e.g., C41 = 0) , and as a result the elasticity tensor is simplified
to [32]
Ê C11 C12 C12 0 0 0 ˆ
ÁC C11 C12 0 0 0 ˜
Á 12 ˜
ÁC C12 C11 0 0 0 ˜
Ccubic = Á 12 (13.8)
Á 0 0 0 C 44 0 0 ˜˜
Á 0 0 0 0 C 44 0 ˜
Á ˜
Ë 0 0 0 0 0 C 44 ¯
However, for crystals that belong to the hexagonal crystal

symmetry group (such as wurtzite GaN compositions), the elasticity
constant is described by four components [8]:
Ê C11 C12 C13 0 0 0 ˆ
ÁC C11 C13 0 0 0 ˜
Á 12 ˜
Á C13 C13 C11 0 0 0 ˜
C hcp =Á ˜ (13.9)
Á 0 0 0 C 44 0 0 ˜
Á 0 0 0 0 C 44 0 ˜
Á ˜
Ë 0 0 0 0 0 (C11 - C12 )/ 2¯
The above matrices are used to find the stress in materials grown
along (0 0 1) and (0 0 0 1) for the case of cubic and hexagonally
Piezoelectricity in Semiconductors 415
symmetric crystals. For all other growth directions, the appropriate

Euler rotation matrix should be applied to find the correct elasticity
tensor. We focused only on these two crystal symmetries because
the highly luminescent semiconductors, as suitable candidates for
laser cooling, fall within these two categories.
It should be noted that the converse piezoelectric effect is the
generation of strain by applying an electric field across a piezoelectric
material. The piezoelectric strain tensor (d) is used to connect these
two physical observables:
d = C–1e (13.10)
where e is the piezoelectric tensor and is described by only one
nonvanishing element for III–V semiconductors [1]:
Ê 0 0 0 e14 0 0ˆ
e = Á0 0 0 0 e14 0˜ (13.11)
Á ˜
ÁË 0 0 0 0 0 e14 ˜¯
whereas for wurtzite crystals, it has the form [43]:
Ê 0 0 0 0 e15 0ˆ
e=Á 0 0 0 e15 0 0˜ (13.12)
Á ˜
ÁË e e31 e33 0 0 0˜¯
31
The above equations form a complete set for finding the

displacement with respect to the applied voltage to a piezoelectric
material and vice versa. In the next sections, we describe how the
separameters can be used to coherently control the lattice vibrations
(damping or amplification) in a piezoelectric QW structure. In this
regard, we note that there are many reports about the generation
of the acoustic phonons in piezoelectric heterostructures by
optical excitation. For instance, lattice-mismatched GaN/InGaN QW
structures possess a giant piezoelectric field. Upon absorbing short
optical pulses by this material, electron–hole pairs are generated,
which travel across the multiple–quantum well (MQW) region due
to the built-in piezoelectric field. The motion of the photogenerated
carriers screen the piezoelectric field and induce a mechanical strain
wave and acoustic phonons [6, 7, 24, 39]. The generation of the
acoustic phonons alters the refractive index of the material, which
manifests itself in the transmission modulation of the probe beam,
as shown in Fig. 13.2.
Figure 13.2 Transmission modulation versus time for a 2.2 nm/7 nm

In0.1Ga0.9N/GaN MQW due to the presence of acoustic phonons.
Reprinted with permission from Ref. [7]. Copyright 2004, AIP
Publishing LLC.
We will investigate the opposite of the described process to

coherently absorb the acoustic phonons and achieve microscopic
laser cooling of piezoelectric heterostructures. This coherent
process is based on optomechanical backaction cooling concepts
which are the subject of the next section.
13.3 Basics of Optomechanical Cooling and

Amplification
The interaction of the light and mater can result in an efficient
energy transfer between the mechanical and the optical fields.
There are many mechanisms that can be utilized to achieve an
efficient energy transfer. In this section, we only focus on backaction
optomechanics and we will use the same concept to demonstrate the
cooling of large number of phonon modes in a bulk solid. Backaction
dynamical interaction can be realized if there is an optical and
mechanical resonator and both are coupled to each other. During
the recent years many miniature nano- and micro-optomechanical
Basics of Optomechanical Cooling and Amplification 417
resonators have been fabricated that made the observation of this

effect possible. The potential applications include ultrasensitive
motion and force sensors used in gravitational wave detectors [3],
photonic clocks [18], high-precision accelerometers, atomic force
microscopy, and parametric amplification.
Movable mirror
x
Optical cavity
Figure 13.3 Schematics of the system suitable for optomechanical
backaction interaction. It consists of a movable mirror (the
displacement of the mirror is shown by x) and resonating
optical field that is coupled to the mechanical state of the
mirror through the optical cavity.
Consider an optical cavity with one movable mirror, as shown in

Fig. 13.3. The coupled mechanical and optical equations of motion
can be written as follows [22]:
da 1Ê 1 1 ˆ
= iD ( x )a - Á +
1
˜ a+i s (13.13a)
dt 2 Ë t 0 t ex ¯ t ex
d2 x Wm Ê dx ˆ Ê 1 ˆ
+ + W2m x = Á (Frp + FL ) (13.13b)
dt2 2Qm ÁË dt ˜¯ Ë meff ˜¯
where x is the position of the movable mirror with respect to
equilibrium, Δ is the detuning of the laser with respect to the cavity
resonance, τ0 is the intrinsic photon decay rate of the cavity, τext
is the extrinsic photon decay rate of the cavity (mainly due to the
mirror loss), Ωm is the mechanical resonance frequency, Qm is the
mechanical quality factor of the mirror, and meff is the effective
mass of the mirror. |a|2 is the optical energy stored in the cavity and
|s|2 is the optical power incident on the cavity. Fep is the force due
to radiation and FL is the Langevin force, which is responsible for
thermal vibrations of the mirror. According to the above equations,
one can figure out that the movement of the mirror changes the
detuning (i.e., Δ(x) = Δ + (ω0/L)x, where L is the length of the Fabry–
Perot cavity and ω0 is the resonance radial frequency of the cavity),
which in turn changes the optical power inside the optical resonator.
On the other hand the change in the optical energy of the resonator
results in a coherent change in the optical force exerted on the
moving mirror. As we shall see in the following, with a correct phase
difference between the dynamical optical force and the mechanical
motion, an efficient energy transfer between the optical mode and
the mechanical mode can be achieved. To start this analysis, we note
that the optical force exerted on the mirror wall is equal to 2|a|2/
(cTtr) in which Ttr is the round-trip transit time (~2L/c) and c is the
speed of light in the cavity. We expand |a|2 in the weak retardation
regime, that is, 1/t >> Ωm according to Eq. 13.13a to obtain
Ê t 2t / t ex ˆ È Ê 8Dt 2 ˆ w Ê 8Dt 2 ˆ w 0 dx ˘ 2
F (t ) = Á Í1 + Á 2 2 ˜ R x -t Á 2 2
0
˜ ˙s
Ë cTrt 4D 2t 2 + 1 ˜¯ ÍÎ Ë 4t D + 1 ¯ Ë 4t D + 1 ¯ R dt ˙˚
(13.14)
-1
where t = t 0-1 -1
+ t ex
and the first two terms in this series are the
adiabatic response of the force to the changes in position of the
moving mirror. As the mirror moves, the instantaneous detuning
changes which results in the change of optical power and force as
described by these two terms. The second term corresponds to the
mechanical displacement and hence results in an induced optical
stiffness. The power transfer between the optical mode and the
mechanical mode can be found by evaluating F.dx/dt and averaging
it over one period of mechanical oscillation. By doing so, it turns out
that only the third term in the above equation is responsible for a
nonzero cooling/amplification rate because it is in phase with the
velocity. It is also inferred from the above equation that the sign
of the detuning determines the direction of the energy transfer
between the optical mode and the mechanical mode. A red-detuned
laser leads to cooling of the mechanical vibration that would change
the balance in favor of anti-Stokes radiation, as schematically shown
in Fig. 13.4, and the reverse is true for a blue-detuned laser. In
general in any system based on the backaction mechanism, to have
Basics of Optomechanical Cooling and Amplification 419
an effective cooling or amplification an in-phase force component

with the velocity is required. We note that the force applied by the
photons inside the cavity does not need to be the radiation pressure.
If photons initiate any kind of mechanism that exert force on the
mirror, they can lead to damping/amplification of the mechanical
motion and should be considered in the theory. To have a better
understanding we expand the light-induced forces (Fopt) into its
components:
t
dFn (t)
Fopt = ÂÚ
n 0
dt
gn (t - t )dt (13.15)
where Fn denotes the force exerted by a particular mechanism

activated by the photons and gn is equal to 1 – exp(–t/τn). τn is the
average time for Fn to reach equilibrium after a sudden change in
the position of the moving mirror. By solving the above equation in
the limit of very small displacements (which is the case for thermal
vibrations), one can find the following relation for the position of the
mirror in the Fourier domain:
xw =
ÂF n n ,w
/(1 + iwt n ) + FL ,w w02
(13.16)
K w eff
2
- w 2 + iG eff w
where Fn,ω and FL,ω are the amplitudes of the light-induced force and
the Langevin force at the frequency of ω. τn is the characteristic delay
time of Fn,ω. K is the mechanical rigidity of the mirror (which is equal
to meff W2m). ωeff and Γeff are the effective resonance frequency and
the damping rate of the moving mirror in the presence of the optical
field inside the cavity and can be expressed as follows [26]:
ÊW ˆ È w 0t n Kn ˘
G eff = Á m ˜
Ë Qm ¯
Í1 - Qm
ÍÎ
Âw t
n
2 2
n
˙
+ 1 K ˙˚
(13.17a)
È Kn ˘
Â
1
w eff
2
= w 02 Í1 + ˙ (13.17b)
ÍÎ n
w 2t n2 + 1 K ˙˚
where Kn is the light-induced rigidity and is equal to –∂Fn, 0/∂x, where
Fn,0 is the static light-induced force at equilibrium. Considering the
mechanical dispersion of the moving mirror and spectral properties
of the thermal driving force, one can obtain the following formula for
amplitude of vibration between ω – δω/2 and ω + δω/2:
2kBT w 02G
<xw >2 = dw (13.18)
p K (w eff
2
- w 2 )2 + (G eff w )2
Intensity Δ1 Δ2
AS
S
S
AS
ω0 ω
ω1 ω2
Figure 13.4 The resonator dispersion and stokes (S) and anti-Stokes
(AS) components of the photons in the cavity. Two different
conditions with negative laser detuning (ω1) and positive laser
detuning (ω2) are shown. In the first scenario the anti-Stokes
component is larger than the Stokes component, leading to the
cooling of the mechanical mode. The reverse is true for positive
detuning, which results in the amplification of the mechanical
motion.
By defining the temperature (Teff) according to equipartition

Ê1
theorem for a 1D harmonic oscillator Á K < x 2> = kBTeff , where kB
Ë2
is the Boltzmann constant) and finding the spectral average of the
square of the displacement < x2 >) using Eq. 13.18, one can show that
w 02 G
Teff = T (13.19)
w eff
2 G eff
The above relation shows that both the effective resonance
frequency (modified by the induced optical rigidity) and the
effective damping rates are responsible for the change in
temperature. According to Eq. 13.17 it can be shown that for Kn =
–K and t n = 2 / w 0, maximum cooling with minimum temperature
Coulomb Interaction in Semiconductors and Its Application for Laser Cooling 421
of Teff = 4 2T / Qm can be achieved. The delay (τn) is necessary to

provide the phase shift required to transfer the energy between the
mechanical and the optical modes, whereas the stiffness increases
with the magnitude of the optical force exerted on the mirror as it
moves out of its equilibrium condition.
From the above discussion, it can be inferred that to have an ef-
fective optomechanical cooling not only a backaction force with cor-
rect phase and large magnitude is required, but also the mechani-
cal quality factor should be high. Extension of the above concepts to
phonon modes of a bulk material is possible and forms the basis of
the Coulomb-assisted piezoelectric cooling, as described in Section
13.6. In a bulk material, each phonon mode can be considered as a
mechanical mode with a low quality factor that leads to a low cool-
ing efficiency. However, once the cooling mechanism is activated on
all phonon modes of the material, a microscopic drop of tempera-
ture is expected. Similar to optomechanical cooling systems, where
the minimum achievable temperature is set by the photo-induced
heating, in bulk semiconductors parasitic absorption of pump and
PL limit the cooling process.
13.4 Coulomb Interaction in Semiconductors

and Its Application for Laser Cooling
Coulomb interaction of carriers here refers to the electrostatic
attraction between negative and positive charges in a host medium
that results in the change of the optoelectronic properties of the
material. We will show that these effects are more pronounced
in confined 2D (QWs) and 3D (quantum dot systems) because of
shorter distances between the electrons and holes. In this regard, the
excitonic effect has always been the focus of research and has been
one of the most important implications of the Coulomb interaction
in semiconductors. The excitons are the photoexcited electrons and
holes bound by the Coulomb attractive force. They are similar to
excited states of a hydrogen atom, as opposed to free electrons and
holes that resemble ionized hydrogen atoms. The excitonic effect is
most apparent at low temperatures where the scattering rate with
high energy phonons is not high enough to exceed the orbiting rate
of the exciton and dissociate it effectively. On the other hand, at

high temperatures, the optical phonon broadening is higher which
prevent the excitons from resolving in the absorption spectrum of
the material. Nonetheless, in the last three decades, the advancement
of epitaxial semiconductor growth has led to ultrafine MQWs, which
are suitable for harvesting the benefits of excitons, even at room
temperature. The reason is that in such confined systems, the radius
of the exciton is much smaller along the growth directions, resulting
in the reduction of the exciton radius in the lateral directions. This
is due to the fact that the energy of the excitons is minimized when
the exciton has a spherical shape. When the dimension of an exciton
reduces, its binding energy and its orbiting rate increase, and hence
the exciton dissociation rate by scattering with high energy phonons
decreases. As a result, a more favorable situation for the presence of
excitons in lower-dimensional systems is provided.
Previous theories have shown that the excitonic effect is quite
beneficial for laser cooling of solids. This is due to the enhanced
excitonic absorption of the pump laser below the bandgap energy,
and also it is proven beneficial by the amplified radiative decay of
the excitons. Rupper et al. [37, 38] has conducted a comprehensive
theoretical investigation to find the emission and absorption spectra
of bulk GaAs with the presence of excitonic effects. According
to that study, the enhancement in excitonic absorption leads to
enhancement of radiative decay at higher energies. Furthermore,
the Coulomb-induced steepening of the absorption tail helps to have
significant absorption of the photons at low energies in order to
overcome the adverse effects of the background absorption of the
input laser.
Confined systems with delocalized electron and hole wave
functions are of particular interest for implementing the Coulomb
effects. In such systems, increasing the density of electrons and holes
leads to an effective electric field between the particles that either
tilts the energy bands or screens the external electric field. Due to
the quantum-confined Stark effect (QCSE), the red shift or blue shift
of bandgap follows respectively. This dynamical effect leads to an
effective nonlinearity that can be utilized for the purpose of laser
cooling of solids. One particular scenario is shown schematically in
Fig. 13.5. The population of electrons and holes screens the external
electric field and blue shifts the transition energy. Therefore, after
Coulomb Interaction in Semiconductors and Its Application for Laser Cooling 423
the carriers reach thermal equilibrium, not only do they gain the
thermal energy of kT but also they experience an inherent blue shift
due to the Coulomb interaction. As a result, the difference between
the energy of the spontaneously emitted photons and the absorbed
photons increases, which could be an advantage, according to
Eqs. 13.1–13.3, for increasing the laser cooling efficiency. Similar
behavior has been demonstrated in type II QW structures, where
the excitation of free carriers leads to a blue shift of radiation by the
specific type II band alignment [10]. This type of band alignment
plays the same role as a built-in electric field and separates the
excited electron and hole pairs. This separation directly affects the
Auger and radiative recombination rates. In the next section, the
experimental and theoretical investigations to clarify these effects
are outlined.
e-
ΔEe
Barrier Well Barrier
ΔEh
h+
Figure 13.5 Schematics of a quantum well structure with an external

applied bias. The electron and hole pairs are delocalized. The
population of these delocalized states results in screening of
the external electric field. Therefore at equilibrium the tilt in
the energy band decreases and electrons and holes get closer
to each other, as shown by the dashed lines. The reduction
of the effective electric field within the quantum well is
associated with the blue shift of electron and hole energy
levels, as denoted by ΔEe and ΔEh. The blue shift adds to the
thermal blue shift of the radiation kT.
13.5 Effect of the Separated Electron and Hole

Wave Functions on the Recombination
Rates
As explained earlier, confined structures with delocalized electron–
hole wave functions can exhibit a blue shift of radiation by means
of the Coulomb interaction between the photogenerated carriers. In
this section, we briefly discuss the radiative and Auger recombination
in such systems. These two parameters play a major role in the
external quantum efficiency and the laser cooling efficiency (see
Eqs. 13.1–13.3) and should be investigated in detail.
The radiative recombination rate at thermal equilibrium without
any photoexcitation is basically equal to the black-body radiation
rate, and is also equal to the absorption rate of the material. This
ensures that the temporal variations of the carrier densities and the
number of photons within the material are zero. For a material with
a bandgap of Eg, we can write [20]
8p c
•
(hn )2
(
a (hn ) n0 (hn ) )
(hc )3 EÚ
2
Bni2 = d(hn ) (13.20)
Ê hn ˆ
exp Á -1
Ë k T ˜¯
g
B
where B is the radiative recombination coefficient, ni is the intrinsic
carrier density, α(hν) and n0(hν) are the absorption and the refractive
index, respectively, of the material at a photon frequency of ν, and h
is Planck’s constant. By simplifying the above equation, we obtain
8p c
B= a g ng2(E g + kBT )2 kBT exp( -E g /kBT ) (13.21)
ni2(hc )3
where αg and ng are the absorption and the refractive index at
the average PL wavelength. The above equation shows that at
thermal equilibrium, the absorption and radiative recombination
are proportional to each other. For a QW structure, the interband
absorption coefficient, neglecting the excitonic effects, has been
found to be
a (hn ) = C Â Â
2 2 2
Inm e .pcv d (Enm (k|| ) - hn )( fvm - fcn )
n ,m
V k
||
(13.22)
Effect of the Separated Electron and Hole Wave Functions on the Recombination Rates 425
where C is a constant of proportionality. The internal summation is

taken over all in-plane wavevectors (k||) and the external summation
is taken over all electron and hole energy levels (Enm). V is the volume
of the crystal, ê is the unit vector characterizing the polarization of
the incident electromagnetic field, and pcv is the momentum matrix
element of the corresponding electron and hole states. f denotes the
Fermi distribution of the carriers. One can notice that the absorption
is proportional to the square of the overlap integral of the electron
and hole wave functions (Inm). Since the radiative recombination
coefficient is directly proportional to the absorption, the radiative
recombination rate also increases with the square of the overlap
integral. It is quite evident that the separation of the electron and
holes results in the decrease the absorption and the radiative
recombination of the material. Now let’s see the consequence of
electron–hole pair separation on the Auger recombination rate of
the QW structures. In general, the behavior of Auger recombination
is substantially different in bulk and heterostructures. The latter
shows nonthreshold Auger properties that have power–law
temperature dependence [19, 20] as opposed to bulk materials with
Auger recombination rates that exponentially increase with the
temperature [21, 22].
As mentioned earlier in Section 13.1, Auger recombination
involves Coulomb scattering between three carriers. In MQW
structures, the CHCC Auger recombination is dominant [20]. In
this process, one electron and one hole recombine and excite
another electron in the conduction band to a higher energy level.
The third electron finally thermalizes with the phonon bath, that is,
it relaxes to a lower energy level converting its energy to heat. In
heterostructures, the momentum conservation is violated along the
direction of growth leading to the nonthreshold behavior of the Auger
recombination. The suppressed nonthreshold Auger recombination
has been observed in type II structures with separated electron
and hole wave functions [27]. Here we also provide experimental
and theoretical evidences for type I QW systems to show that the
nonthreshold Auger is in fact the circumstance for any MQW system
with delocalized electron and hole wave functions.
One period of the grown MQW structure is shown in Fig. 13.6. It
consist of a 3 nm InGaAs layer for confining the lowest energy state
of the heavy holes and a 18 nm InGaAsP layer with the bandgap of

1.43 μm for confining the electrons. Due to the presence of a built-
in electric field equal to 45 kV/cm, the energy bands are tilted, and
the electron and hole wave functions are separated with respect to
each other. There is also a thin (1 nm) InAlAs layer in between the
two layers in order to further separate the electrons and holes. The
electron barrier is the InAlAs layer which is adjacent to the InGaAsP,
and the hole barrier is InP adjacent to InGaAs layer. Both of them
have a thickness of 17 nm. All of the described layers are lattice-
matched to the InP substrate. To simulate these types of structures
both the band lineup and the electron and hole effective masses
are required. The energy band lineup for any composition can be
obtained using the model solid theory [48]. The effective masses
close to the band minima are found using the Luttinger parameters
obtained by applying the k.p method. These data are necessary to
solve the Schrodinger equation self consistently with the Poisson
equation. After finding the energy levels and the wave functions, the
carrier distribution in different energy levels can be obtained using
Fermi–Dirac statistics and the PL spectra can be evaluated. The
results have been plotted and compared to the experimental data in
Fig. 13.7. The close agreement between the two confirms the good
quality of growth and validity of the theoretical calculations.
-5
InAlAs InGaAsP InAlAs
InGaAs InP
-5.5
Energy (eV)
e1
-6
-6.5 hh1
lh1
-7
0 50 100 150 200 250 300 350 400
Position (Å)
Figure 13.6 The band structure and the wave function of the grown
structure used to investigate the Auger recombination process.
The electron wave function is shown by the dotted lines,
whereas the first heavy-hole and light-hole wave functions are
shown by the solid and dashed lines.
Normalized PL intensity
e1-lh
1
0.8 Simulated
0.6 e1-hh
0.4 Measured
0.2
0
1.3 1.35 1.4 1.45
Wavelength (μm)
Figure 13.7 The simulated and measured PL spectra. Two peaks corre-
sponding to the electron–light hole and electron–heavy hole
recombinations were identified and indicated. The simulated
curve is a solid line and the measured curve is shown by the
dotted line.
To find the Auger recombination rate experimentally, one way

is to perform the excitation-dependent PL measurements. In these
experiments, the intensity of the excitation laser is varied and
the detected power of the PL is used to extract the recombination
coefficient [23]. Figure 13.8 shows the measured PL curves. There
are also lines that indicate the slope of the Auger and radiative
recombination decay for typical PL–pump intensity curves of such
materials. According to the slope of the measured curves, it is quite
evident that the intensity of the pump laser was substantially high
to make the Auger mechanism dominant in all of the measurements.
Therefore these data can be used to obtain the Auger coefficient
versus temperature relationship, as shown in Fig. 13.9. The behavior
of the Auger recombination can be explained by considering two
ranges of the temperatures. For low temperatures, the nonthreshold
Auger recombination coefficient is found to be [11]
3.5
ÈE ˘
32 2 U me
2 1
C=Í 0˙ w B I11 (13.23)
3p
ÍÎ E g ˙˚ kBT mhh
kg4
where Eg is the bandgap, ωB is the electron Bohr frequency, E0 is
equal to the electron energy in the first state, and U is the conduction
band offset. The effective electron and heavy-hole masses are

denoted by me and mhh, respectively, and kg = 2me E g / . I11 is
the overlap integral between the first electron and hole levels (n =
m = 1). By evaluation of all of the parameters , a very good fit to
the experimental data can be found as shown in Fig. 13.9. At high
temperatures, however, the carriers occupy higher-energy states
and the confinement probability decreases. As a result, the behavior
of the Auger recombination becomes closer to the bulk Auger
described by [45]
Ê m Eg ˆ
C = C0exp Á - ˜ (13.24)
Ë 1 + m kBT ¯
where μ is ratio of the of the electron to heavy-hole effective mass
and C0 is the constant of proportionality. A good fit using this
formula to the Auger coefficient obtained at high temperatures can
be found as shown in Fig. 13.9. Due to the reduced overlap integral
between the electron and holes, the Auger recombination decreases
significantly as compared to other bulk and QW systems made with
similar material composition (see Table 13.1).
Log10 PL Peak voltage (mV)
2.8
T=77 K
2.6
2.4
2.2
T=300 K Auger Slope
2
0.9 1 1.1 1.2 1.3 1.4
Log10 Pump power (mW)
Figure 13.8 The PL peak voltage versus pump power for different
temperatures ranging from 77 K up to 300 K. The Auger slope is
shown by the dashed line and the radiative slope by the dotted
line. The slopes of the measured curves are close to Auger
process, which shows that the carrier density is high enough
for the Auger process to dominate in all of the measurements.
We also note that the best carrier concentration for the purpose
of laser cooling is equal to A / C , where A is the SRH recombination
rate [42]. Therefore, the decrease in the Auger recombination

coefficient necessitates the use of higher laser powers to achieve
maximum laser cooling efficiency.
Auger coefficient (cm6/s)
Measured
-28 Fitted (non-threshold Auger)
10
Fitted (bulk Auger)
-30
10
100 150 200 250 300

Temperature (K)
Figure 13.9 The extracted Auger recombination coefficient from the curves
in Fig. 13.8. These data are plotted versus the temperature and
fitted by nonthreshold and bulk-type Auger, as described in the
text.
Table 13.1 Comparison of Auger recombination coefficient at T = 200 K

of the introduced structure to the quantum well structures
without electron and hole wave function separation and also
the bulk material with similar composition
Structure (cm6/s)
1.1 Auger coefficient
1.2 The structure-separated electron 6.5 × 10–30

1.3 hole wave function
1.4 Type I In0.53Ga0.47As QW (lattice-matched)a 1.3 × 10–28
1.5 Type I In0.46Ga0.54As QW (tensilely strained layer)b 6 × 10–29
1.6 Type I In0.66Ga0.34As QW (compressively strained layer)c 7 × 10–29
1.7 Bulk InGaAsP (Q1.3)c 2.5 × 1029

aRef. [17]
bRef. [16]
cRef. [40]
13.6 Formalism of Coulomb-Assisted Laser

Cooling of Semiconductors
To facilitate the laser refrigeration of semiconductors, the QCSE can
be used to increase the bandgap after the generation of electron–
hole pairs, as we explained in Section 13.4. This way, the emitted
photons are guaranteed to have a higher energy than the absorbed
ones. The increase of energy level difference of electrons and holes
due to the QCSE happens due to the screening of the internal field
by the photogenerated carriers. The band deformation caused by
photogenerated carriers in asymmetric type II QWs has the same
origin. Although we provide a theoretical basis to show that the
Coulomb interaction can lead to cooling in piezoelectric materials,
the physical interoperation of the energy transfer in this process in
nonpiezoelectric materials is still under investigation. In this section,
we shall present the most relevant phenomena that may reveal the
physics behind this phenomenon in near the future.
The first question that comes to mind is the location of the
electrons and holes immediately after creation. According to Refs.
[47, 46] in which the dynamic solution of self-consistent Poisson
and Schrödinger equations is presented, the electron–hole pairs
immediately occupy the empty wave functions (wave functions
calculated without the presence of photogenerated carriers);
therefore the screening immediately happens but it decreases over
time until the self-consistent state is reached. The bandgap increases
immediately upon creation of electron–hole pairs and the origin of
this energy shift is a question that should be addressed in the future.
Anti-Stokes cooling requires a high enough photon extraction ef-
ficiency as well as quantum efficiency. Therefore the implementa-
tion of this technique is extremely difficult for bulk semiconductors.
The Coulomb interaction between photogenerated electron–hole
pairs in strong piezoelectric materials such as GaN/InGaN QWs can
assist anti-Stokes laser cooling. By comparing to cavity backaction
mechanism [26], we explain how cooling or heating depends upon
laser detuning with respect to the bandgap. Therefore, this method
can also be also utilized for coherent phonon generation. The pro-
cess is illustrated in Fig. 13.10a and can be explained as follows:
The propagation of phonons with polarization along the growth
Formalism of Coulomb-Assisted Laser Cooling of Semiconductors 431
direction induces piezoelectric field, which coherently changes the

bandgap by the QCSE. The change of bandgap results in the change
of photogenerated carrier density. On the other hand, the electric
field between the electron–hole pairs due to spatial charge sepa-
ration is dependent upon carrier density. Therefore, the change of
carrier density leads to the change of the electric field between the
electron–hole pairs and the associated piezostress on the atoms. In
this cycle, a backaction mechanism exists: the strain caused by the
phonons results in a coherent piezostress due to the Coulomb poten-
tial of electron–hole pairs. Therefore, there will be a nonzero power
transfer between the phonons and the carriers. This process could
lead to net cooling if the energy gained by the carriers is extracted
efficiently from the material by radiative recombination, as shown in
Fig. 13.10b.
The strain induced by the phonons, generates a piezoelectric
field, which through the QCSE changes the bandgap:
Êd Ê (
Ê 2C 2 ˆ ˆ C Ê uz (0, t ) - uz LQW , t
E g (t ) = CStark Á 31 Á C11 + C12 - Á 13 ˜ ˜ 33 Á
) ˆ˜
Ë  Ë Ë C33 ¯ ¯ C13 ÁË LQW ˜¯
(13.25)
where uz is the out-of-plane displacement of atoms within the
width of the QW structure (LQW) and the Cs are elastic stiffness
coefficients. To obtain the above formula, first the in-plane strain
C
component (εxx) was obtained by multiplying the factor - 33
2C13
Ê u (0, t ) - uz ( LQW , t ) ˆ
to the average out of plane strain Á z ˜ [52]. The
Ë LQW ¯
in-plane strain component corresponds to a piezoelectric field
Ê 2d ˆ Ê 2C 2 ˆ
with a magnitude equal to Á 31 ˜ Á C11 + C12 - 13 ˜  xx , where
Ë  ¯Ë C33 ¯
d31 is the piezoelectric constant and ε is the electric permittivity
of the medium [52]. The piezoelectric field changes the bandgap
through the QCSE and this change is calculated by including
the Stark coefficient (CStark). It should be noted that for the QW
structure described in here the change of bandgap by this process
is much larger than that of deformation potential (ΔEg,d), that is,
DE g Ê 2C d ˆÊ 2C 2 ˆ
= Á Stark 31 ˜ Á C11 + C12 - 13 ˜ >~ 10 where D1 and D2 are
DE g,d Ë  ( D2 + D4) ¯ Ë C33 ¯
deformation potential parameters.
(a) Phonon
dynamics
Change of Phonon
piezo strain strain
Change of internal Piezoelectric

electric field field
Coulomb
QCSE
interacon Change
Between e—-h+ Of photo-
generaon
(b) Absorbed Laser power
Phonon scaering
Electron
Thermal Phonon LO
Bath
reservoir Bath AC Equilibrium at Te
Equilibrium at TL Coulomb
interacon
Non-Radiative
recombination Radiave
recombinaon
Thermal generaon
Figure 13.10 (a) The diagram showing the processes that lead to the
Coulomb backaction mechanism and (b) the energy flow
schematics. The Coulomb cooling increases the acoustic
phonon absorption, as shown by the red flash.
In a piezoelectric material, the electrostatic field between

the spatially separated electrons and holes leads to an effective
piezoelectric stress. It is known that the piezoelectric stress can be
related to the electrostatic field using the piezoelectric coefficient
[7, 33]: e33λ21n, in which e33 is the piezoelectric coefficient and λ21
is the Coulomb strength parameter, which is the ratio of the induced
electric field to the carrier density (n). Therefore the change in
bandgap (as predicted by Eq. 13.25) leads to the change of carrier
density and the piezoelectric stress (σpiezo). To find the change in
carrier density we use the rate equation
dn
= G - ( An + Bn2 + Cn3 ) (13.26)
dt
where A, B, and C are the SRH, radiative, and Auger recombination
rates, respectively. G is the generation rate and is calculated as
follows:
a 2PA LQW I ˆ
I(1 - exp( -a ( w - E g )LQW ) +
Iabs 1 + a 2PA LQW I ˜¯
G(w ) = =
LQW w LQW w
(13.27)
where, Iabs is the intensity of the absorbed photons, LQW is the length
of the QW where the carriers are confined, α is the absorption
coefficient that depends on detuning (ћω – Eg), I is the laser intensity,
Eg is the bandgap, and α2PA is the two-photon absorption coefficient.
Assuming small changes in carrier density we can write the small
signal form of Eq. 13.27 as follows:
d Dn
= DG - RDn (13.28)
dt
where Δ means the change in the quantity of interest and R = A +
2Bn + 3Cn2. By substituting Δn = Re[ΔN(ωq)eiwqt] (ωq being the radial
frequency of the phonon), we find the following equation for ΔN(ωq):
Ê ˆ
Á ˜
Á 1 Ê iw q R ˆ ˜ Ê DG(w q ) ˆ
N(w q ) = Á -Á 2 ˜ (13.29)
Ê wq ˆ
2 ÁË R + w q2 ˜¯ ˜ ÁË R ˜¯
Á ˜
Á 1 + ÁË R ˜¯ ˜
Ë ¯
and consequently
R
Dn (t ) =
1
Dn0 - d Dn0 /dt (13.30)
Ê wq ˆ
2
R + w q2
2
1+Á ˜
Ë R ¯
d Dn
where Δn0 is the solution of the rate equation when DG . By
dn dt
using Dn0 = DE g,piezo , we can write
dE g
1 Ê dn ˆ R Ê dn ˆ Ê d DE g,piezo ˆ
Dn(t ) = Á ˜ DE g,piezo - Á ˜Á ˜
Ê w q ˆ Ë dE g ¯ R2 + w q2 Ë dE g ¯ Ë dt ¯
2
1+Á ˜
Ë R ¯
(13.31)
Therefore the component of the piezostress, which is out of
phase with atomic displacement due to phonons, will be equal to
1 Ê dn ˆ
s piezo,o = e33 l21 Á ˜ DE g (13.32)
Ê w q ˆ Ë dE g ¯
2
1+Á ˜
Ë R ¯
It is then possible to calculate the cooling intensity from the

coherent piezoelectric stress:
Le-h
Ê du ˆ Ê 1ˆ Ê duz ˆ
Icooling,q = Â < s piezo Á z ˜ > = Á ˜
Ë dt ¯ Ë a¯ Ú <s piezo Á
Ë dt ˜¯
> dx
0
(13.33)
in which the cooling intensity for each atomic layer is calculated
by taking an average in the time domain (as denoted by <> signs)
and the results from all atomic layers within the average electron–
hole separation length (Le-h) has been summed up. The summation
was replaced by the integration assuming that the electric field
due to charge separation is uniform within the active region and
is negligible elsewhere. Using Eqs. 13.25, 13.32, and 13.33 and
assuming that uz = uzm cos(kq.r – ωqt), where uzm is the amplitude of
vibration perpendicular to QW layers and kq is the wavevector of the
phonon, we can obtain the following equation for cooling per unit
area of mode q:
ÊL ˆ
Icooling,q = 8 Á e-h ˜
Ë arV ¯
1 Ê 1 ˆ Ê 1 ˆ Ê kqz ˆ Ê 1ˆ
CCoulomb ÁË v ˜¯ ÁÁ 2 2 ˜˜ Á k ˜ (1 - cos(kqz Le- h ))ÁË nq + 2 ˜¯ w q
s Ë Le-h kqz ¯ Ë q ¯
2
Ê wq ˆ
1+Á ˜
Ë R ¯
(13.34)
where ρ is the mass density, a is the lattice constant, Le-h is the aver-
age separation of electron and hole wave functions, V is the volume
of the crystal, vs is the group velocity of the phonon mode, kq and kqz
are the amplitudes of the phonon wavevector along the direction of
the propagation and along the crystal growth direction, respectively,
and nq is the phonon occupation number of mode q. CCoulomb is an ef-
fective elasticity and is equal to the ratio of backaction piezoelectric
stress to the average phonon’s strain along the direction of epitaxial
growth and is expressed as
Ê 1ˆ Êd Ê Ê 2C 2 ˆ ˆ C Ê dn ˆ
CCoulomb = Á ˜ e33 l21CStark Á 31 Á C11 + C12 - Á 13 ˜ ˜ 33 Á ˜
Ë 2¯ Ë  Ë Ë C33 ¯ ¯ C13 Ë dE g ¯
(13.35)
The obtained formula for cooling per unit area (Eq. 13.34)
shows that the mode with an out-of-plane wavelength of 2.33Le-h
experiences maximum cooling. In Fig. 13.11 the phenomena that
lead to Coulomb-assisted cooling are shown as a function of time for
this particular mode. The change of bandgap and the carrier density
due to out-of-plane strain of the phonon leads to a coherent change
of piezostress, as shown in Fig. 13.11a–d. The component of the
piezostress, which is 180° out of phase with carrier velocity, results
in the net cooling power (Fig. 13.11d–f). The total cooling per unit
area extracted from the phonons by means of Coulomb interaction
is obtained by summing up the contribution of all phonon modes.
The cooling per unit area of phonons modes with a specific energy
2p k||q dk||q k zq
of ћvs(kq)kq is equal to Id3k = Icooling ,q ¥ 3
, where L is
Ê 2p ˆ
ÁË L ˜¯
the length of the crystal and k||q is the magnitude of the in-plane
wavevector. Thus summing up Id3k over the whole Brillion zone
Ê Ê 2p ˆ Ê p pˆ
Á k||q ŒÁ 0, ˜ and k zq ŒÁ - , ˜ [34] gives the net cooling per
Ë Ë a ¯ Ë a a¯
unit area:
p Ê 2p ˆ
kzq = k||q =Á ˜
a Ë a ¯
Ê 2 ˆ
Icooling =
p
ÚÚ Á 2 ˜
Ë ap r ¯
kzq =- ,k||q =0
a
Ê 1 ˆ1 Ê 1ˆ
CCoulomb Á ˜ (1 - cos(kqz Le-h ))ÁË nq + ˜¯ k||qdk||qdk zq
Ê w q ˆ Ë Le-h kqz ¯
2
2
1+Á ˜
Ë R¯
(13.36)
(a) εzz~ uz(0,t)-uz(LQW ,t) (b) ΔEg~Δεzz
Δεzz,max 0
ΔEg,max 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
(c) Δncoh (d) Δσo~Δno

Δncoh = Δni +Δno
Δno,max 0
Δσo,max 0
0 2 4 6 8 10 12 14 16 18 20
0 2 4 6 8 10 12 14 16 18 20
(e) (f)
Δvavg 0
Pcooling 0
0.23 0.57 0.91 t(ps)

0.23 0.57 0.91 t(ps)
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Figure 13.11 The time diagram of (a) the average out-of-plane strain, (b) the
change of bandgap due to the change of piezoelectric field, and
(c) the change of carrier density due to the change of bandgap.
It is composed of an in-phase (Δni) and an out-of-phase (Δno)
component. The amplitude of the out-of-phase component is
5.4 × 10–5 times lower than the total carrier density change
for the phonon mode under consideration. (d) The piezostress
due to Δno and (e) the average velocity. It is in phase with the
piezostress shown in (d) and therefore results in a net cooling
power, as shown in (f).
We can then simplify the cooling per unit area by defining the
function F:
p Ê 2p ˆ
kqz = ,k||q =Á ˜
a Ë a ¯
Ê 1 ˆ
ÚÚ
1
= Á ˜
v2s (kqz
2
+ k||2q ) Ë Le -hkqz ¯
kqz =0 ,k||q =0 1+
R2
Ê ˆ
Á ˜
Á ˜
Á ˜
1 1˜
(1 - cos(kqz Le-h ))Á + k||qdk||qdk zq
Á
Á
Ê v
Á s
(k 2
qz + k||2q ) ˆ
˜
2˜
˜
Á exp Á
kT ˜ -1 ˜
Á ÁË ˜¯ ˜
Ë ¯
(13.37)
and then express Icooling as
Ê 2 ˆ
Icooling = Á 2 ˜ CCoulomb  (13.38)
Ë ap r ¯
Therefore the cooling efficiency can be calculated as
Ê ˆ
Á ˜
Icooling Á 2 ˜
hcooling = =Á Ccoulomb 
Iabs ˆ˜
Á ap 2 rI(1 - exp( -a L ) + a 2PA LQW I ˜
ÁË QW
1 + a 2PA LQW I ˜¯ ˜¯
(13.39)
A wurtzite c-GaN/InGaN QW, as shown in Fig. 13.12, was
considered to explore the Coulomb effects. The k.p method in
conjunction with Poisson equation can be used to calculate the
self-consistent energy levels and the absorption of the MQW [52]
in presence of Coulomb interaction. Effective masses and band line-
ups were calculated according to Ref. [9]. The QW structure consists
of 1 nm of In0.15Ga0.85N sandwiched between 6 nm of GaN barriers.
As can be seen in Fig. 13.12, the spatial separation of electron and
hole wave functions due to the presence of a large piezoelectric
field, leads to the screening of the built-in and piezoelectric electric
field after photogeneration. As a result, significant blue shift of the
interband transition energies is expected for a large photogenerated
carrier density due to the QCSE.
-3
GaN In0.15Ga0.85N GaN
-3.5
-4
Energy (eV)
3.212
-4.5
Ee-Ehh (eV)
3.21 Phonon’s stress
-5
3.208
-5.5
3.206
1 2 3 4
-6 x 10
19
Piezo stress
n(cm-3)
-6.5
-7
-7.5
0 20 40 60 80 100 120 140
Position (Å)
Figure 13.12 The band diagram of the designed MQW system. The electron
wave function is in green, whereas the heavy-hole wave
function is in red color. The inset shows the bandgap versus
carrier density. Note that the direction of piezostress will be
reversed for a phonon with opposite strain. This results in an
effective phonon damping.
The inset of Fig. 13.12 shows the bandgap versus the carrier
density with the slope equal to Le–hλ21. According to Fig. 13.13a, we
can see that the net cooling efficiency increases significantly with the
presence of the Coulomb interaction. We can see that the net cooling
is possible for a laser intensity of 5 mW/μm2 and for the detunings
between 0–4 THz. The dependence of the net cooling efficiency on
laser intensity can be explained by the fact that both the external
quantum efficiency ηext and the Coulomb cooling are dependent on
laser intensity. At very high intensities, the probability of two-photon
absorption increases, which is almost independent of detuning near
the band edge [23] . As a result, in this regime, the change of carrier
density with the change of bandgap reduces and the Coulomb
cooling efficiency is suppressed. This behavior can be understood
considering that the Coulomb cooling efficiency decreases due to
the reduction of dn/dEg. For extremely low excitation levels, the SRH
recombination of the photogenerated carriers reduces the external
quantum efficiency and the net cooling. The reduction of the internal
quantum efficiency for large blue detunings is due to the increased
carrier density and domination of the Auger recombination rate. In
both cases, the reduction of the internal quantum efficiency effectively
diminishes the positive effects of the Coulomb interaction on the net
laser cooling. As a result, below and above certain detunings, the net
cooling efficiencies for both cases (with and without the Coulomb
effect) become essentially equal.
(a) cooling
0
With Coulomb
-20 Without Coulomb
ηc,net (%)
5î 10 3 mW/μm2
-40
5 mW/μm2
-60
-80 0.5 mW/μm2
-100
-20 -15 -10 -5 0 5 10 15 20
Detuning (THz)
(b)
-4
10
only anti-Stokes
anti-Stokes with
Coulomb cooling
-6
τnrb (s)
10
30 meV
-8
τnr=7.8×10-8 s 19 meV
10
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ηe
Figure 13.13 (a) The net cooling efficiency versus laser detuning for various
laser intensities. Cases with both anti-Stokes and Coulomb
cooling are shown by the dotted lines. The solid lines show only
anti-Stokes cooling. These simulations have been performed
for the structure that has an absorption broadening of 19 meV
when the entire PL is extracted out from the material without
reabsorption. (b) The break-even nonradiative recombination
lifetime (τnrb) versus extraction efficiency for only anti-Stokes
cooling (solid line) and both the anti-Stokes and Coulomb
cooling (different values of absorption broadening has been
considered). The 30 meV absorption broadening is a usual
experimental value [19]. The lowest broadening is expected to
be close to 19 meV (the value for GaN epilayers [15]). The red
line represents the nonradiative recombination lifetime for a
similar structure [53]. In all of the simulations, it is assumed
that the temperature is 300 K. Reprinted with permission from
Ref. [30]. Copyright 2014, AIP Publishing LLC.
The break-even nonradiative lifetime is defined as the shortest

nonradiative lifetime that allows laser cooling (see Eqs. 13.1 and
13.2) [41]. The presence of the internal Coulomb cooling lowers the
break-even nonradiative lifetime, as demonstrated in Fig. 13.13b.
This improvement becomes more pronounced when the absorption
broadening is lower since dn/dEg increases as the absorption
broadening decreases. By considering the nonradiative lifetime in
a similar InGaN/GaN QW structure [53], we see that net cooling is
possible with the presence of Coulomb cooling when the broadening
is 30 meV and the extraction efficiency is better than 85% [53]. In
this structure, the composition of the InGaN layer is close to the GaN
barrier. Therefore the ultimate value of the broadening (19 meV)
for the GaN epilayer [15], was also used to recalculate the break-
even nonradiative lifetime, as shown in Fig. 13.13b . In this case,
the net cooling is possible when the extraction efficiency is better
than 55%. This is well below the reported extraction efficiency of
73% reported for light-emitting diodes (LEDs) made out of these
materials, indicating the feasibility of this method of cooling in the
near future.
To conclude this section, it was demonstrated that the Coulomb
interaction could be utilized to facilitate the laser cooling of
piezoelectric materials. The process is similar to optomechanical
backaction cooling, where there is a backaction force due to
displacement. Here the backaction force comes from the piezoelectric
effect when the atoms move with phonons. The Coulomb cooling
efficiency is highly dependent on laser intensity and detuning. It is
shown that at optimum conditions the Coulomb interaction could
make the net cooling possible for the structure under consideration.
It has been also shown that the reduction of absorption broadening
enhances the Coulomb cooling efficiency because of a higher change
of the carrier density per change of carrier detuning. Increasing the
laser intensity increases the Coulomb cooling efficiency until the
point where the two-photon absorption dominates. However, the
optimum laser intensity not only depends on the Coulomb cooling
efficiency but also on the recombination processes. As for the laser
detuning, the Coulomb cooling efficiency is maximized near the band
edge where the slope of absorption versus detuning is the highest.
Owing to the high efficiency of this mechanism, the next step is to
Comparison of Coulomb-Assisted Cooling in Piezoelectric Materials 441
investigate it in other strong piezoelectric semiconductors such as

III–V quantum dots.
13.7 Comparison of Coulomb-Assisted Cooling

in Piezoelectric Materials with Collisionally
Aided Laser Cooling
In this section, we briefly introduce the laser cooling by means
of collision redistribution of radiation in gaseous systems. Our
description of this type of cooling is mainly based on the paper
published by Berman and Stenholm [4] and the experimental
demonstration of the cooling reported by Vogl and Weitz [49]. This
topic is covered in detail in the Chapter 1 on collisional redistribution
laser cooling in this book. We compare this method of laser cooling
with the Coulomb-assisted laser cooling of piezoelectric materials.
Consider a compressed gaseous system, as shown in Fig. 13.14a.
Without any collision between the atoms and neglecting the
Doppler shift of the radiation, the luminescence linewidth is very
narrow and the energy of the radiated photons is essentially equal
to the energy of the photons belonging to the excitation laser. A
red-detuned excitation of an atom, far from its natural linewidth,
results in excitation of virtual carriers [21]. However, when the
atoms collide with each other, the “virtually excited” atoms can
gain energy out of the collisions and can get really excited to the
upper state. As a result the luminescence energy will be close to
the bandgap of the atom and is spectrally broadened because of
collisions as shown in Fig. 13.14b. The simple picture described in
here neglects the linewidth broadening by the Doppler effect (which
is equal to w D ª w ex /c kT /M , where ωex is the radial frequency of
the excitation photons and M is the mass of the atom). Furthermore
certain conditions need to be met in order to have an effective
cooling/heating by this process. Assuming that the total thermal
energy of the particle to be ET ≈ kT , the duration of the collision and
the collision rate are represented by τc and Γ, respectively, then for a
certain detuning (Δ), we should have [4]:
ET
> t c-1 > D > G (13.40)

The cooling efficiency is related to the difference between the

frequency of the emitted (vfl) and the absorbed photons(vab):
n fl - n ab D
hcooling = a (vab ) = a (vab ) (13.41)
n ab n ab
where α(vab) denotes the absorbance of the pumped photons.
Therefore to increase the cooling or heating efficiency, the detuning
should be increased but not to the extent to result in a considerable
decrease of absorbance. Also the condition of τc|Δ| < 1, as indicated
in Eq. 13.40, allows one to analyze the system using “impact
approximation” whereas the condition ET t > 1 ensures the thermal
c
equilibrium of the atomic ensemble with the effective temperature
of T. Each collision excites the atom to its upper state and results
in an energy gain equal to the laser detuning. Considering that the
spontaneous decay of the atoms is dominant (which is the case when
the population inversion is very small), then the excitation rate can
be obtained as
2 c 2G 2 c 2G
n2 = @ 2 (13.42)
D + (G + g /2)
2 2
D
where γ is the radiative rate of the atoms. At the steady state, that is,
n◊ 2 = 0, the occupation probability of the excited level is equal to
2c 2G
n2 = (13.43)
D 2g
The condition of |Δ|> Γ, as indicated in Eq. 13.40, is therefore
necessary to have a small excitation and dominant radiative decay of
the atoms (λ >> n◊ 2). The normalized energy change per unit time is
expressed by
dET /ET D 2c 2G
= (13.44)
ET ET D 2
It is possible to find the cooling power and efficiency using the
above equations. The cooling power can be found as follows:
P = γn (E – E ) – n◊ E = n◊ (ħΔ)
cool 2 2 1 2 ex 2 (13.45)
where E1, E2, and Eex are the energies of the first and second levels
and the photons of the excitation laser. For the absorbed power, we
can write
P = n◊ E
abs 2 ex (13.46)
Comparison of Coulomb-Assisted Cooling in Piezoelectric Materials 443
(a) Collision (b)
Broadened excited
Level due to collisions
Quartz E2
window
Excitaon Excitaon energy

laser
fluorescence
E1
Carrier fluorescent
atoms atoms
Figure 13.14 (a) The schematics of a gaseous system that consists of active
florescent atoms and carrier atoms. The collision between
the atoms results in broadening of the luminescence and
upconversion of the luminescence energy with the respect to
the absorbed photon energy. (b) The energy level diagram and
the excitation and radiation of an atom in the system. With
collisions the atoms can be excited to the upper state and emit
photons with a center frequency equal to (E2 – E1)/ħ. Therefore
a red-detuned excitation results in cooling.
Therefore the cooling efficiency is equal to Pcool/Pabs = ħΔ/Eex.

This method of cooling was experimentally demonstrated by Vogl
and Weitz, who provided a simple explanation for the cooling
mechanism, as shown by the graph of Fig. 13.15. the bandgap of
the atoms depends on the interatomic distance, the bandgap has
a minima and a maxima. Since the pump laser is red tuned with
respect to the bandgap, therefore the absorption of the pump laser
can only happen when the atoms get close enough to each other to
have a reduced bandgap. However, on average the distance between
the atoms corresponds to a higher bandgap. Since the fluorescence
is a slower phenomenon compared to the collision events, atoms
emit photons with a higher energy equal to their bandgap energy
at the average interatomic distance. Neglecting the fluorescence
trapping, in this process the energy of collisions gets extracted by
the fluorescence.
It is worthwhile to compare this method of cooling with that of
Coulomb-assisted cooling, as described in the previous section. In
the latter approach the propagation of the phonon through the QW
structure changes the average interatomic distance and results in a

net strain. Similar to collisionally aided cooling, this effect results in
an effective change of the MQW bandgap. The change of bandgap
results in an immediate change of the photogenerated carrier
density in phase with the average atoms’ velocity. Therefore, the
corresponding piezoelectric force on the atoms does a net work on
them. This work can either coherently retard or amplify the atomic
vibration by the phonon mode. In the case of cooling, the excess
energy taken from the phonon bath is electrostatically transferred
to the photogenerated electrons through the piezoelectric effect.
Similar to the collisionally aided cooling, the fluorescence happens
at a much slower timescale compared to cooling/heating cycle of
the phonons and therefore the extra energy of the carriers can be
extracted out from the material effectively.
Energy
Excited
Fluorescence
excitaon
Ground
d0 Inter-atomic
distance
Figure 13.15 The energies of the ground state and the excited states versus
the interatomic distance. The average interatomic distance
is denoted by d0. This graph is similar to the one indicated
in Ref. [49]. This technique of cooling is similar to Coulomb-
assisted cooling of piezoelectric materials in the sense that
in both cases the photoexcitation happens as the interatomic
distance decreases and the fluorescence is a slower process
that extracts the excess energy of the carriers outside of the
material. However, the mechanism of a shrinking bandgap and
the excess energy gained by the carriers is different for both
cases.
Concluding Remarks 445
It should be noted that the laser is red-detuned from the bandgap

to achieve Coulomb-assisted laser cooling. This is due to the change
of absorption per change of the bandgap for a slightly red-detuned
laser. As a result when atoms get closer to each other, the bandgap
shrinks, as described by Eq. 13.25, and a significant photoexcitation
happens. This is exactly similar to collisionally aided cooling, as shown
in Fig. 13.15, where the reduction in interatomic distance results in
photoexcitation. The excess photogenerated carriers gain an average
energy by the described coherent piezoelectric interaction with the
phonon. Therefore the average energy of the luminescence will be
higher that the absorbed photons and extraction of the PL results in
cooling.

In this chapter we discussed the mechanisms that can potentially
increase the laser cooling efficiency to make it practical for
bulk semiconductors. Our focus was on Coulomb interaction
in piezoelectric semiconductors. The equations regarding the
anti-Stokes laser cooling efficiency and the external quantum
efficiency were investigated to determine different absorption
and recombination parameters and their roles in cooling of solids
with lasers. These parameters include parasitic absorption by
impurities, defects and surface states, extraction efficiency of the PL,
and radiative, Auger, and SRH recombination rates. In addition the
cooling efficiency depends on the bandgap of the material and the
carrier density. It is concluded from the literature that laser cooling
is very weak, even at optimum conditions, rendering it impossible
for bulk semiconductors with the highest material quality achieved
so far. Therefore other mechanisms that can initiate the laser cooling
process should be investigated in order to assist anti-Stokes laser
cooling. To establish the theoretical bases for understanding the
Coulomb-assisted laser cooling, we described both the Coulomb
interaction and the piezoelectricity separately. The Coulomb
interaction here refers to the electrostatic attraction between the
electron and hole pairs. The excitonic effect is a good example, which
has been already shown to increase the laser cooling efficiency due
to the enhancement in the excitonic radiative decay and its effect on
steepening of the band-tail absorption. In general, the electric field

between the photogenerated electron and hole pairs can either blue-
shift or red-shift the transition energies. This phenomenon can be
used for laser cooling purposes, as we have shown in this chapter.
We provided a brief theory of piezoelectricity, particularly for zinc
blende and wurtzite crystals that are currently at the center of
attention for semiconductor laser cooling due to their high internal
quantum efficiency. Our approach is inspired by the optomechanical
cooling of micro- and nanomechanical systems, and therefore we
explained the basics of backaction optomechanics and proceeded to
establish the described theory by using the same mechanism since
each phonon mode acts like a mechanical mode of vibration with low
quality factor. The low quality factor of the phonon modes might be
discouraging at first, however, there exist numerous phonon modes
and the net cooling is the sum of the cooling powers of all those
phonon modes. To validate the assumptions, we considered highly
piezoelectric InGaN/GaN QW systems. Each acoustic phonon mode
changes the bandgap by producing a net strain across the QW region.
Since the material is piezoelectric, the strain is associated with an
effective electric field that changes the bandgap by the QCSE. The
change of the bandgap changes the photogeneration of the carriers.
Now, since the electron and hole pairs are separated in space, the
change in the photogenerated carrier density results in a change in
the electrostatic field between the electrons and holes. The change
in the electric field is associated with the piezoelectric force exerted
on the atoms within the QW region. Therefore there is a closed loop
that starts with the strain produced by the phonons and ends with
a coherent, retarding piezoelectric force that prevents atoms from
vibrating. Using the standard equations regarding photogeneration,
the QCSE, and piezoelectricity, the laser cooling efficiency and the
break-even nonradiative recombination rate can be found with and
without the Coulomb effect. The results show a significant reduction
of the break-even SRH lifetime and reveal the possibility of cooling in
such systems. Finally, we compared this method to the collisionally
aided laser cooling (for detailed information of this method, the
reader is referred to the chapter on collisional redistribution laser
cooling in this book) . In both methods, the change of interatomic
distance results in a change of the bandgap and photoexcitation
occurs with lower energy photons, while the slower radiative
References 447
recombination extracts the excess energy of the material through

emitting higher-energy photons.
It is worth mentioning that many of the proposed ideas can
be further improved or investigated in other material systems.
Furthermore, future experiments are needed to investigate each
of the proposed mechanisms individually. And lastly, although the
above theory applies to piezoelectric materials, the backaction
optomechanical concepts can be used to damp the lattice vibration
with any type of force associated with the photogenerated
carriers. For these reasons, we believe that the extension of the
optomechanical concepts to laser cooling of bulk materials merits
further investigations.
Acknowledgments
The authors would like to acknowledge partial support from NSF
awards nos. ECCS-1310620, ECCS-1206155, and ECCS-0901855, as
well as ARO award nos. W911NF-13-1-0485 and W911NF-11-1-0390.
The authors would also like to acknowledge the help of Travis
Hamilton, Eric Dexheimer, and Abigail Turay in preparation of this
chapter.
References
1. Adachi, S. (1994). GaAs and Related Materials: Bulk Semiconducting

and Superlattice Properties (World Scientific).
2. Arnau, A. (2008). Piezoelectric Transducers and Applications (Springer).
3. Aspelmeyer, M., Meystre, P., Schwab, K. (2012). Quantum
optomechanics, Phys. Today, 65, pp. 29–35.
4. Berman, P. R., Stenholm, S. (1978). Heating or cooling collisionally
aided fluorescence, Opt. Commun., 24, pp. 155–157.
5. Bernardini, F., Fiorentini, V. (2002). First-principles calculation of the
piezoelectric tensor d of III-V nitrides, Appl. Phys. Lett., 80, p. 4145.
6. Chern, G.-W/, Lin, K.-H/, Huang, Y.-K., Sun, C.-K. (2003). Spectral
analysis of high-harmonic coherent acoustic phonons in piezoelectric
semiconductor multiple quantum wells, Phys. Rev. B, 67, p. 121303.
7. Chern, G.-W., Lin, K.-H., Sun, C.-K. (2004). Transmission of light through
quantum heterostructures modulated by coherent acoustic phonons, J.
Appl. Phys., 95, pp. 1114–1121.
8. Christmas, U. M., Andreev, A., Faux, D. (2005). Calculation of electric

field and optical transitions in InGaN∕ GaN quantum wells, J. Appl. Phys.,
98, p. 073522.
9. Chuang, S. L. (1995). Physics of Optoelectronic Devices (Wiley).
10. Dowd, P., Braun, W., Smith, D. J., Ryu, C., Guo, C.-Z., Chen, S., Koelle, U.,
Johnson, S., Zhang, Y.-H. (1999). Long wavelength (1.3 and 1.5 μm)
photoluminescence from InGaAs/GaPAsSb quantum wells grown on
GaAs, Appl. Phys. Lett., 75, pp. 1267–1269.
11. Dyakonov, M., Kachorovskii, V. Y. (1994). Nonthreshold Auger
recombination in quantum wells, Phys. Rev. B, 49, p. 17130.
Nature, 377, pp. 500–503.
13. Fathipour, V., Jang, S., Hassaninia, I., Mohseni, H. (2014). Approaching
high temperature photon counting with electron-injection detectors,
Proc. SPIE, 9220, p. 92200J.
14. Fathipour, V., Memis, O. G., Jang, S. J., Brown, R. L., Nia, I. H., Mohseni, H.
(2014). Isolated electron injection detectors with high gain and record
low dark current at telecom wavelength, IEEE J. Sel. Topics Quantum
Electron., 20, pp. 65–70.
15. Fischer, A., Shan, W., Park, G., Song, J., Kim, D., Yee, D., Horning, R.,
Goldenberg, B. (1997). Femtosecond four-wave-mixing studies of
nearly homogeneously broadened excitons in GaN, Phys. Rev. B, 56, p.
1077.
16. Fuchs, G., Schiedel, C., Hangleiter, A., Härle, V., Scholz, F. (1993). Auger
recombination in strained and unstrained InGaAs/InGaAsP multiple
quantum‐well lasers, Appl. Phys. Lett., 62, pp. 396–398.
17. Hausser, S., Fuchs, G., Hangleiter, A., Streubel, K., Tsang, W. (1990).
Auger recombination in bulk and quantum well InGaAs, Appl. Phys.
Lett., 56, pp. 913–915.
18. Hossein-Zadeh, M., Vahala, K. J. (2010). An optomechanical oscillator
on a silicon chip, IEEE J. Sel. Topics Quantum Electron., 16, pp. 276–287.
19. Jiang, H., Singh, J. (1999). Linear electro-optic effect due to the built-
in electric field in InGaN/GaN quantum wells, Appl. Phys. Lett., 75, pp.
1932–1934.
20. Johnson, S., Ding, D., Wang, J.-B., Yu, S.-Q., Zhang, Y.-H. (2007). Excitation
dependent photoluminescence measurements of the nonradiative
lifetime and quantum efficiency in GaAs, J. Vac. Sci. Technol. B, 25, pp.
1077–1082.
References 449
21. Khurgin, J. (1992). Coulomb enhancement of ultrafast nonlinearities in

quantum-well structures, J. Opt. Soc. Am. B, 9, pp. 157–172.
22. Kippenberg, T. J., Vahala, K. J. (2007). Cavity opto-mechanics, Opt.
Express, 15, pp. 17172–17205.
23. Krishnamurthy, S., Nashold, K., Sher, A. (2000). Two-photon absorption
in GaN, GaInN, and GaAlN alloys, Appl. Phys. Lett., 77, pp. 355–357.
24. Lin, K.-H., Lai, C.-M., Pan, C.-C., Chyi, J.-I., Shi, J.-W., Sun, S.-Z., Chang, C.-
F., Sun, C.-K. (2007). Spatial manipulation of nanoacoustic waves with
nanoscale spot sizes, Nat. Nanotechnol., 2, pp. 704–708.
25. Melgaard, S., Albrecht, A. R., Hehlen, M., Seletskiy, D., Sheik-Bahae,
M. (2014). Optical refrigeration cools below 100 K, Optical Society of
America, paper FTh4D.4.
26. Metzger, C. H., Karrai, K. (2004). Cavity cooling of a microlever, Nature,
432, pp. 1002–1005.
27. Mohseni, H., Litvinov, V., Razeghi, M. (1998). Interface-induced
suppression of the Auger recombination in type-II InAs/GaSb
superlattices, Phys. Rev. B, 58, p. 15378.
28. Nia, I. H., Memis, O. G., Kohoutek, J., Gelfand, R. M., Mohseni, H. (2012).
Surface plasmon enhancement of photon extraction efficiency by silver
nanoparticles: with applications in laser cooling of semiconductors,
Proc. SPIE, 8457, p. 84573V.
29. Nia, I. H., Mohseni, H. (2014). Exploring Coulomb interaction in
piezoelectric materials for assisting the laser cooling of solids, Proc.
SPIE, 9000, p. 90000M.
30. Nia, I. H., Mohseni, H. (2014). A proposal for Coulomb assisted laser
cooling of piezoelectric semiconductors, Appl. Phys. Lett., 105, p.
042102.
31. Nia, I. H., Mohseni, H. (2015). Photoluminescence study of suspended
MQW structures for laser refrigeration purposes, Proc. SPIE, 9380, p.
93800E.
32. Norris, A. N. (2006). Optimal orientation of anisotropic solids, Q. J.
Mech. Appl. Math., 59, pp. 29–53.
33. Babilotte, P., Ruello, P., Vaudel, G., Pezeril, T., Mounier, D., Breteau, J.-
M., Gusev, V. (2010). Picosecond acoustics in p-doped piezoelectric
semiconductors, Appl. Phys. Lett., 97, p. 174103.
34. Pokatilov, E. P., Nika, D. L., Balandin, A. A. (2003). Phonon spectrum
and group velocities in AlN/GaN/AlN and related heterostructures,
Superlattice Microst., 33, pp. 155–171.

Lumineszenz-und Temperaturstrahlung, Z. Phys., 57, pp. 739–746.
36. Ruan, X., Kaviany, M. (2007). Advances in laser cooling of solids, J. Heat
Trans., 129, pp. 3–10.
37. Rupper, G., Kwong, N., Binder, R. (2006). Large excitonic enhancement
of optical refrigeration in semiconductors, Phys. Rev. Lett., 97, p.
117401.
38. Rupper, G., Kwong, N., Binder, R. (2007). Optical refrigeration of GaAs:
theoretical study, Phys. Rev. B, 76, p. 245203.
39. Sanders, G., Stanton, C., Kim, C. S. (2001). Theory of coherent acoustic
phonons in InxGa1−xN/GaN multiple quantum wells, Phys. Rev. B, 64, p.
235316.
40. Sermage, B., Heritage, J., Dutta, N. (1985). Temperature dependence
of carrier lifetime and Auger recombination in 1.3 μm InGaAsP, J. Appl.
Phys., 57, pp. 5443–5449.
41. Sheik-Bahae, M., Epstein, R. I. (2004). Can laser light cool
semiconductors?, Phys. Rev. Lett., 92, p. 247403.
42. Sheik-Bahae, M., Imangholi, B., Hasselbeck, M., Epstein, R., Kurtz, S.
(2006). Advances in laser cooling of semiconductors, Proc. SPIE, 6115,
p. 611518.
43. Shi, C., Asbeck, P. M., Edward, T. Y. (1999). Piezoelectric polarization
associated with dislocations in wurtzite GaN, Appl. Phys. Lett., 74, pp.
573–575.
44. Shimada, K. (2006). First-principles determination of piezoelectric
stress and strain constants of Wurtzite III–V nitrides, Jpn. J. Appl. Phys.,
45, p. L358.
45. Smith, C., Abram, R., Burt, M. (1985). Theory of Auger recombination in
a quantum well heterostructure, Superlattice Microst., 1, pp. 119–123.
46. Turchinovich, D., Jepsen, P. U., Monozon, B., Koch, M., Lahmann, S.,
Rossow, U., Hangleiter, A. (2003). Ultrafast polarization dynamics in
biased quantum wells under strong femtosecond optical excitation,
Phys. Rev. B, 68, p. 241307.
47. Turchinovich, D., Monozon, B., Jepsen, P. U. (2006). Role of dynamical
screening in excitation kinetics of biased quantum wells: nonlinear
absorption and ultrabroadband terahertz emission, J. Appl. Phys., 99,
p. 013510.
48. Van de Walle, C. G. (1989). Band lineups and deformation potentials in
the model-solid theory, Phys. Rev. B, 39, p. 1871.
References 451
49. Vogl, U., Weitz, M. (2009). Laser cooling by collisional redistribution of

radiation, Nature, 461, pp. 70–73.
50. Yoo, K., Albrecht, J., Ram-Mohan, L. (2010). Strain in layered zinc
blende and wurtzite semiconductor structures grown along arbitrary
crystallographic directions, Am. J. Phys., 78, pp. 589–597.
52. Zhao, H., Arif, R., Ee, Y.-K., Tansu, N. (2009). Self-consistent analysis of
strain-compensated InGaN–AlGaN quantum wells for lasers and light-
emitting diodes, IEEE J. Quantum Electron, 45, pp. 66–78.
53. Zhao, H., Liu, G., Zhang, J., Poplawsky, J. D., Dierolf, V., Tansu, N. (2011).
Approaches for high internal quantum efficiency green InGaN light-
emitting diodes with large overlap quantum wells, Opt. Express, 19, pp.
A991–A1007.

Laser Cooling: Fundamental Properties and Applications

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Laser Cooling: Fundamental Properties and Applications

Încărcat de

Drepturi de autor:

Formate disponibile

Laser Cooling

Pan Stanford Series on Renewable Energy — Volume 2

The Rise of Modern Wind Energy

British Library Cataloguing-in-Publication Data

Laser Cooling: Fundamental Properties and Applications

Printed in the USA

1. Laser Cooling of Dense Gases by Collisional

2. Laser Cooling in Rare Earth–Doped Glasses and Crystals 37

2.2.2 Conversion Efficiency 43

2.6.2.1 Photothermal deflection 70

3. Progress toward Laser Cooling of Thulium-Doped

3.3.3.2 Noncontact measurements 105

4. Laser Cooling of Solids around 2.07 Microns:

5. Optically Cooled Lasers 147

6. Methods for Laser Cooling of Solids 189

7. Deep Laser Cooling of Rare Earth–Doped Crystals

8. Bulk Cooling Efficiency Measurements of Yb-Doped

9. Interferometric Measurement of Laser-Induced

10. Fluoride Glasses and Fibers 305

10.4.3 Mechanical Properties 328

11. Crystal Growth of Fluoride Single Crystals for

11.5.4 Sample Fabrication 373

12. Microscopic Theory of Optical Refrigeration of

13. Coulomb-Assisted Laser Cooling of Piezoelectric

Refrigeration or cooling is a process in which a substance is

be laser cooled with anti-Stokes fluorescence. In that case a system

advantages of different host materials with low phonon energy

ipate in multi-phonon relaxation of excited ions. It is demonstrated

structural quality need to be satisfied in order to achieve net bulk

Laser Cooling of Dense Gases by

Anne Saβ, Stavros Christopoulos, and Martin Weitz

Laser Cooling: Fundamental Properties and Applications

experimentally realized in 1981 [11], this technique leads nowadays

subsequently present the experimental setup and the high-pressure

1.2 Redistribution of Radiation

1.2.1 Basic Principle

Figure 1.1 Process of redistribution of radiation for a two-level atom A

It can be shown experimentally that the intensity of the

1.2.2 Cooling Principle

lifetime of the excited alkali state. Our main experimental focus

For small interatomic distances, a photon can be absorbed

Our choice of rubidium atoms as the optically active species

mv12 mv22 mv2 mv22

1.3 Experimental Setup and Methods

1.3.1 High-Pressure Cells

Figure 1.3 Schematic of a high-pressure cell (homemade design). The

The second design is commercial, manufactured by SITEC-Sieber

up to 720 K can be achieved using both cell designs. A manometer,

1.3.2 Spectroscopic Setup

Figure 1.4 Fluorescence detection setup. The incident Ti:sapphire laser

1.3.3 Thermal Deflection Spectroscopy

Regarding Eq. 1.9, we can further simplify dn/dT in order

To get a rough estimation for the analysis of the measured data,

This assumption holds for small temperature changes compared

where r denotes the transverse displacement between cooling and

analysis of the measured deflection data. In our present model, heat

Q(r , z ) µ a ◊ Pcool ◊ e -2r /w02

Inspection of Eq. 1.16 indicates that the temperature change

Figure 1.6 Experimental cooling setup incorporated with thermal deflec-

We discuss results on thermal deflection spectroscopy

1.4 Laser Cooling Experiments on Dense

1.4.1 Fluorescence of a Dense Rubidium–Argon Mixture

Figure 1.7 Recorded fluorescence of rubidium atoms subject to 160 bar

The resonances are also shifted to higher wavelengths, λ(D1¢)